Neuroimaging Biomarkers in Alzheimer's Disease

Thematic Collections
in Biomedical Research
Neuroimaging
biomarkers
inAlzheimer's
disease.
Dr. Samuel Barrack
Published by:
Internet Medical Publishing
Find more titles on our site: www.imedpub.com
Thematic Collections
in Biomedical Research
Neuroimaging
biomarkers
inAlzheimer's
disease.
Dr. Samuel Barrack
Title: Neuroimaging biomarkers

in Alzheimers disease
Autor: Samuel Barrack
ISBN-13: 978-1-49227-442-1
ISBN-10: 1492274429
Cover design and Layout: [email protected]

Maquetacin: David Mrquez
[email protected]

Publisher: Internet Medical Publishing
[email protected]
http://imedpub.com/
First edition: 2013
Disclaimer: This collection is based in articles published by the Public Library of Science (PLOS). PLOS applies the
Creative Commons Attribution License (CC-BY) to works we publish. Under this license, authors retain ownership of
the copyright for their content, but allow anyone to download, reuse, reprint, modify, distribute, and/or copy the
content as long as the original authors and source are cited.
II
Neuroimaging biomarkers in Alzheimers disease
Preface for Biomarkers for

Alzheimers disease
In view of the growing prevalence of AD worldwide, there is an urgent
need for the development of better diagnostic tools and more effective
therapeutic interventions. Indeed, much work in this field has been done
during last decades. As such, a major goal of current clinical research in AD
is to improve early detection of disease and presymptomatic detection of
neuronal dysfunction, concurrently with the development of better tools
to assess disease progression in this group of disorders. All these putative
correlates are commonly referred to as AD-related biomarkers. The ideal
biomarker should be easy to quantify and measure, reproducible, not
subject to wide variation in the general population and unaffected by comorbid factors. For evaluation of therapies, a biomarker needs to change
linearly with disease progression and closely correlate with established
clinico-pathological parameters of the disease.
There is growing evidence that the use of biomarkers will increase our
ability to better indentify the underlying biology of AD, especially in its early
stages. These biomarkers will improve the detection of the patients suitable
for research studies and drug trials, and they will contribute to a better
management of the disease in the clinical practice. Indeed, much work in
this field has been done during last decades. The vast number of important
applications, combined with the untamed diversity of already identified
biomarkers, show that there is a pressing need to structure the research
made on AD biomarkers into a solid, comprehensive and easy to use tool to
de deployed in clinical settings.
To date there are few publications compiling results on this topic. That is why
when I was asked to address this task I accepted inmediately. I am happy
to present you a bundle of the best articles published about biomarkers for
Alzheimers disease in recent times.
Signed
Dr Samuel Barrack
III
IV
Contents
Neuroimaging
biomarkers: MRI
Diffusion Tensor Metrics as Biomarkers
in Alzheimers Disease
Neuroimaging
biomarkers: PET
2
Cost-Effectiveness of Magnetic
Resonance Imaging with a New
Contrast Agent for the Early Diagnosis
of Alzheimers Disease
16
Multi-Method Analysis of MRI Images
in Early Diagnostics of Alzheimers
Disease
30
CSF and Brain Structural Imaging
Markers of the Alzheimers
Pathological
Cascade
Sensitivity and Specificity of Medial
Temporal Lobe Visual Ratings
and Multivariate Regional MRI
Classification in Alzheimers Disease
MRI Markers for Mild Cognitive
Impairment: Comparisons between
White Matter Integrity and Gray
Matter Volume Measurements
65
Diagnosed Mild Cognitive Impairment

Due to Alzheimers Disease with
PET Biomarkers of Beta Amyloid
and Neuronal Dysfunction
66
Regional Amyloid Deposition in
Amnestic Mild Cognitive Impairment
and Alzheimers Disease Evaluated
by [18F]AV-45 Positron Emission
Tomography in Chinese Population
72
Multi-Method Analysis of MRI Images

in Early Diagnostics of Alzheimers
Disease
80
39
46
Combined Evaluation of FDG-PET

and MRI Improves Detection
and Differentiation of Dementia
89
Plasma Based Markers of [11C]

PiB-PET Brain Amyloid Burden
97
Generative FDG-PET and MRI Model

of Aging and Disease Progression
in Alzheimers Disease
107
55
Evaluating Alzheimers Disease
Progression Using Rate of Regional
Hippocampal Atrophy
118
Diffusion Tensor Metrics as Biomarkers in Alzheimers

Disease
Julio Acosta-Cabronero1*, Stephanie Alley2, Guy B. Williams3, George Pengas1, Peter J. Nestor1,4
1 Cognition, Memory and Language Group, Neurology Unit, Department of Clinical Neurosciences, School of Clinical Medicine, University of Cambridge, Cambridge,
United Kingdom, 2 Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences, University of Cambridge, Cambridge, United Kingdom,
3 Wolfson Brain Imaging Centre, Department of Clinical Neurosciences, School of Clinical Medicine, University of Cambridge, Addenbrookes Hospital, Cambridge, United
Kingdom, 4 German Center for Neurodegenerative Diseases (DZNE), Magdeburg, Germany
Abstract
Background: Although diffusion tensor imaging has been a major research focus for Alzheimers disease in recent years, it
remains unclear whether it has sufficient stability to have biomarker potential. To date, frequently inconsistent results have
been reported, though lack of standardisation in acquisition and analysis make such discrepancies difficult to interpret.
There is also, at present, little knowledge of how the biometric properties of diffusion tensor imaging might evolve in the
course of Alzheimers disease.
Methods: The biomarker question was addressed in this study by adopting a standardised protocol both for the whole
brain (tract-based spatial statistics), and for a region of interest: the midline corpus callosum. In order to study the evolution
of tensor changes, cross-sectional data from very mild (N = 21) and mild (N = 22) Alzheimers disease patients were examined
as well as a longitudinal cohort (N = 16) that had been rescanned at 12 months.
Findings and Significance: The results revealed that increased axial and mean diffusivity are the first abnormalities to occur
and that the first region to develop such significant differences was mesial parietal/splenial white matter; these metrics,
however, remained relatively static with advancing disease indicating they are suitable as state-specific markers. In
contrast, increased radial diffusivity, and therefore decreased fractional anisotropythough less detectable earlybecame
increasingly abnormal with disease progression, and, in the splenium of the corpus callosum, correlated significantly with
dementia severity; these metrics therefore appear stage-specific and would be ideal for monitoring disease progression. In
addition, the cross-sectional and longitudinal analyses showed that the progressive abnormalities in radial diffusivity and
fractional anisotropy always occurred in areas that had first shown an increase in axial and mean diffusivity. Given that the
former two metrics correlate with dementia severity, but the latter two did not, it would appear that increased axial
diffusivity represents an upstream event that precedes neuronal loss.
Citation: Acosta-Cabronero J, Alley S, Williams GB, Pengas G, Nestor PJ (2012) Diffusion Tensor Metrics as Biomarkers in Alzheimers Disease. PLoS ONE 7(11):
e49072. doi:10.1371/journal.pone.0049072
Editor: Wang Zhan, University of Maryland, College Park, United States of America
Received June 21, 2012; Accepted October 4, 2012; Published November 7, 2012
Copyright: 2012 Acosta-Cabronero et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Medical Research Council, UK (to PJN); National Institute for Health Research, Cambridge Biomedical Research Centre, UK (to PJN); and Alzheimers
Research UK (to JAC and GP). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: [email protected]
Introduction
measures, can remain relatively unperturbed. Conflicting results,

however, have been reported in Alzheimers disease; some studies
show emphatic mean diffusivity (MD) differences, believed to be
largely driven by l1 alterations [5,7,8]; some report stronger RD
effects [7,9,10,11]the former in limbic tracts only; and some
show,
in
addition,
highly-abnormal
FA
behaviours
[2,4,7,10,12,13,14]. While these findings have helped define the
landscape of diffusion changes in Alzheimers disease, it is unclear
how the various tensor metrics evolve over time, and, whether
better understanding of such evolution may explain some of these
apparently conflicting results. In order to address this issue, we
used a common acquisition protocol to examine the evolution of
tensor changes over the course of Alzheimers disease by studying
both cross-sectional data at differing dementia severities and
longitudinal change over a 12 month period. We performed
whole-brain analyses and also assessed a directly-visualised white
matter tract that is known to be severely damaged in Alzheimers
There is presently considerable interest in trying to expedite

therapeutic development through the use of biomarkers to track
change in Alzheimers disease. To date, most biomarker work with
magnetic resonance imaging (MRI) has focused on structural
acquisitions to measure atrophy [1]. A potential strength of MRI,
compared, for instance, to nuclear medicine imaging techniques, is
that multiple types of dataoffering complimentary information
can be acquired in a single scanning session. Diffusion tensor
imaging (DTI) is one such method that offers information about
white matter integrity. Early work using DTI in Alzheimers
disease focused particularly on fractional anisotropy or FA [2,3,4],
though some studies have identified that this measure is insensitive
to early white matter disruption in Alzheimers disease [5,6]. This
is because both axial (l1) and radial diffusivity (RD) increase and
therefore FA, which is a function of the ratio of these two
PLOS ONE | www.plosone.org
November 2012 | Volume 7 | Issue 11 | e49072

Diffusion Tensor Biomarkers in AD
Table 1. Demographic summary including cognitive features for Alzheimers disease patients and for a group of elderly controls.
General Demographics
Global Cognition
ACE-R Subscores
Control
(N = 26)
Alzheimers
Disease (N = 43)
Very mild Alzheimers

disease (N = 21)
Mild Alzheimers
disease (N = 22)
Gender, M:F
11:15
26:17
13:8
13:9
Age at imaging, years
68 (6)
70 (6)
72 (5)
69 (6)
MMSE/30
29.1 (0.8)
23.7 (3.6)**
25.9 (1.6)**
21.7 (3.9)**
ACE-R/100
94.6 (3.0)
71.5 (11.9)**
81.4 (4.0)**
62.1 (8.8)**
Attention & Orientation/18
17.9 (0.3)
15.4 (2.7)**
17.0 (1.4)*
13.9 (2.8)**
Memory/26
24.5 (1.9)
10.9 (4.3)**
13.7 (4.1)**
8.2 (2.4)**
Fluency/14
12.1 (1.7)
8.3 (3.2)**
10.6 (1.9)*
6.2 (2.6)**
Language/26
24.9 (0.9)
23.0 (2.6)**
24.6 (1.2)
21.5 (2.7)**
Visuospatial/16
15.4 (0.9)
13.7 (2.8)**
15.1 (1.2)
12.3 (3.1)**
Disease severity (as measured by ACE-R) enabled a median split of the patient cohort into very mild and mild Alzheimers disease subgroups.
Where appropriate, group values are given as mean (SD).
MMSE/30 = Mini-mental state examination score out of 30-point total; ACE-R/100 = Addenbrookes cognitive examination-revised score out of 100-point total.
Wilcoxon rank-sum significance levels: *P,0.01 (Alzheimers disease worse than controls, two-tailed); **P,0.05 (Alzheimers disease worse than controls, two-tailed,
Bonferroni-corrected on n = 28 tests); P,0.05 (Mild worse than very mild Alzheimers disease, two-tailed, Bonferroni-corrected on n = 28).
doi:10.1371/journal.pone.0049072.t001
disease i.e. the corpus callosum [3,4,7,8,11,12,13,15,16]. The aim

was to identify whether the various tensor metrics might show
differential preference as biomarkers to track change or for early
diagnosis, and if so, which are the best for each purpose.
Table 2. Cognitive profile evolution of a group of early-stage

Alzheimers disease patients that was followed-up for a period
of 12 months.
Methods
Alzheimers Disease
(Longitudinal, N = 16)
Ethics Statement
General
Demographics
The study was approved by the Cambridgeshire 2 Research

Ethics Committee (Reference: 07/H0308/215) and by the
National Hospital for Neurology and Neurosurgery & Institute
of Neurology Joint Research Ethics Committee (Reference: 05/
Q0512/12).
Written informed consent was obtained from all the participants. In the context of this study, although the target population
were patients suffering from a degenerative brain condition, we
expected them to have capacity to consent as only the mild stages
of Alzheimers disease were studied. Before inclusion, every patient
was assessed by an expert cognitive neurologist to ensure capacity
and this was indeed the case. Although we had ethical permission
to scan patients who lacked capacity (with caregiver consent), this
was not needed in the present study.
Global Cognition MMSE/30

ACE-R/100
12 Months
8:8
25.1 (2.0)
22.6 (4.9)1
77.5 (7.3)
70.3 (13.6)**
ACE-R Subscores Attention &

16.1 (1.1)
Orientation/18
14.8 (3.5)
Memory/26
13.5 (3.9)
Fluency/14
9.6 (2.5)
9.8 (3.4)**
8.5 (3.8)
Language/26
23.6 (1.8)
23.1 (2.5)
Visuospatial/16 14.7 (1.5)
14.0 (2.9)
Cognitive measures are given as mean (SD).

MMSE/30 = Mini-mental state examination score out of 30-point total; ACE-R/
100 = Addenbrookes cognitive examination-revised score out of 100-point
total.
Wilcoxon signed-rank significance levels: 10.01,P,0.05 (Alzheimers disease: 12
months worse than baseline, two-tailed); **P,0.05 (Alzheimers disease: 12
months worse than baseline, two-tailed, Bonferroni-corrected on n = 7).
Subjects
Cross-sectional cohorts. Forty-three patients with earlystage probable Alzheimers disease according to criteria from the
National Institute of Neurological and Communicative Disorders
and Stroke and the Alzheimers Disease and Related Disorders
Association (NINCDS-ADRDA) [17] were recruited from the
memory clinic at Addenbrookes Hospital (Cambridge, UK). For
cross-sectional comparisons, 26 matched controls were also
recruited and were screened to exclude neurological or major

psychiatric illness. They performed normally on cognitive screening: mini-mental state examination or MMSE [18] and Addenbrookes cognitive examinationrevised or ACE-R [19]. The
control exclusion criteria was ACE-R,88 (out of 100). Note that
25 patients (ACE-R = 69.5612.5) and 12 elderly controls previously assessed [5] were also included in the present study.
Enabled by an ACE-R median split, patients were further
subcategorised according to disease severity into very mild (best
50% ACE-R) and mild Alzheimers disease cohorts (worst 50%
ACE-R) (Table 1). The very mild patient group (N = 21) included
16 subjects who were diagnosed with the mild-cognitive impairment stage of Alzheimers diseasei.e. these were patients scanned
with mild-cognitive impairment that were subsequently shown to
Figure 1. Cognitive status. Depiction of average cognitive profiles

for all subject cohorts assessed in this study.
doi:10.1371/journal.pone.0049072.g001
Gender, M:F
Baseline
Imaging
have probable Alzheimers disease by confirming progressive

cognitive decline with longitudinal follow-up. Although significantly impaired overall on both global cognitive measures, the
very mild Alzheimers disease group was unimpaired on language
and visuospatial subsections of the ACE-R at the time of scanning.
The mild Alzheimers disease group, however, was impaired in all
subdomains of the ACE-R compared both to controls and to the
very mild group of patients.
Longitudinal cohort. A subgroup of 16 Alzheimers disease
subjects 9 of which were diagnosed with mild cognitive
impairment at baseline was followed-up with scans and
neuropsychology tests taking place 12 months apart. Serendipitously, the cognitive scores shown in Table 2 revealed a similar
transition from baseline to 12 months as was seen in the crosssectional data on the very mild Alzheimers disease stage relative to
controls (Table 1)the memory subdomain of the ACE-R was also
the most impaired, followed by statistical trends towards fluency
and attention/orientation deficits; whereas in contrast, language
and visuospatial abilities, despite being slightly reduced, remained
comparatively preserved. Group-average cognitive profiles are
summarised in Figure 1.
All patients were scanned within an average of 1.3 months

(standard deviation = 1.9 months) from a cognitive assessment.
MRI scans were performed on a Siemens Trio 3T system (Siemens
Medical Systems, Erlangen, Germany) with gradient coils capable
of 45 mT/m and 200 T/m/s slew rate. A standard 12-channel
phased-array total imaging matrix head-coil (Siemens Medical
Systems, Erlangen, Germany) was used to transmit/receive radiofrequency signals.
Diffusion tensor imaging. Diffusion datasets were acquired
using a twice-refocused, single-shot, echo-planar imaging pulse
sequence [20]: repetition time (TR)/echo time (TE)/number of
excitations = 7800 ms/90 ms/1; matrix, 96696; 63 contiguous
axial slices; isotropic voxel resolution of 26262 mm3; bandwidth
of 1628 Hz/pixel and echo spacing of 0.72 ms. The tensor was
computed using 63 non-collinear diffusion directions (b = 1000 s/
mm2) that were maximally spread by considering the minimal
energy arrangement of point charges on a sphere, and one scan
without diffusion weighting (b = 0 s/mm2, b0). We allowed for
parallel acquisition of independently-reconstructed images using
generalised, autocalibrating, partially-parallel acquisitions or
GRAPPA [21]; acceleration factor of 2 and 39 reference lines.
The total scan time was 89449.
Volumetric T1 imaging. T1-weighted anatomical images
were also acquired in the same session. The structural scan
consisted of 3D magnetisation-prepared, rapid gradient-echo
(MPRAGE) volumes with the following imaging parameters:
TR/TE/inversion time/flip angle = 2300 ms/2.86 ms/900 ms/
9u,
144
slices,
1926192
matrix
dimensions
and
1.2561.2561.25 mm3 voxel size. Receiver bandwidth and echo
spacing were 240 Hz/pixel and 6.7 ms, respectively. The total
scan time was 79239.
Ultrafast T2 imaging. Whole-brain, T2-weighted, half-Fourier acquisition, single-shot turbo spin echo (HASTE) images were
acquired to ensure that vascular pathology was not significant in
any subject. The following scan parameters were used: TR/TE/
flip angle/turbo factor = 2000 ms/89 ms/150u/205; matrix,
3206256; 25 axial slices (distance factor: 20%); voxel resolution,
0.760.964 mm3; 5/8-phase partial Fourier transform; bandwidth
and echo spacing of 401 Hz/pixel and 5.58 ms, respectively.
GRAPPA mode was enabled with an acceleration factor of 2 and
24 reference lines, resulting in a total scan time of 52 seconds.
In all acquisitions the field of view was aligned in stereotactic
space: the axial plane was aligned to the anterior commissure
posterior commissure line, and the sagittal plane to the interhemispheric fissure. In addition to stereotactic alignment, in order
to maximise acquisition consistency across subjects, the scanning
bed was adjusted to co-localise the centre of the thalamus in the
mid-sagittal plane with the scanner isocentre.
Data Processing and Analysis
Diffusion tensor parametric maps. The Oxford Centre

for Functional MRI of the Brain (FMRIB) software library (FSL
v4.1.2) [22] was used to correct for motion and eddy currents, fit
the diffusion tensor and compute axial, radial and mean diffusivity
as well as fractional anisotropy whole-brain maps. Initially, each
diffusion-weighted volume was affine-aligned to its corresponding
b0 image using the FMRIBs linear image registration tool (FLIRT
v5.4.2) [23]; this pre-processing step corrects for motion artefacts
and eddy-current distortions. In addition, in order to eliminate
spurious voxels, brain masks of each b0 image were generated
using the brain-extraction tool (BET v2.1) [24] with fractional
threshold, f = 0.1, and vertical gradient, g = 0. The FMRIBs
diffusion toolbox (FDT v2.0) was then used to fit the tensor and
Figure 2. Corpus callosum subdivision. Depiction of the semiautomated callosal subdivision into splenium, truncus and genu (top),
and their intersection with the mean FA skeleton inferred from N = 69
subjectsN = 43 Alzheimers disease patients and N = 26 matched
controls (bottom).

Figure 3. Cross-sectional study of very mild Alzheimers disease. TBSS results for the very mild Alzheimers disease group compared to
controls. Statistical maps (thresholded at TFCE-P,0.05) for increased axial/radial diffusivity and reduced FA overlaid onto the mean FA skeleton and
the MNI152 template. Coronal depths are given in millimetres.
compute the diagonal elements (l1, l2 and l3) at each brain voxel,
from which the derived metrics RD, MD and FA were also
inferred. Note that negative primary eigenvalues were deemed
unphysical and were set to 0a visual inspection of the spatial
distribution of negative eigenvalues revealed that they were located
in the periphery of white matter bundles i.e. adjacent to other
tissue types and far from tract centres.
Tract-based spatial statistics (TBSS) analysis. The TBSS
approach [25] was used to perform whole-brain statistical analyses
at white matter tract centres. Spatial normalisation was achieved
by warping all FA images to the 16161 mm3 FMRIB58_FA
standard template (FMRIB, University of Oxford, UK) in
MNI152 space (Montreal Neurological Institute, McGill University, Canada) using the FMRIBs non-linear image registration tool
(FNIRT v1.0). Allpatients (N = 43) and controls (N = 26)warped

FA maps were averaged to create the mean FA template, from
which the mean FA skeleton was derived (FA.0.2). Finally, all
subjects spatially normalised FA, l1, RD and MD data were
projected onto the skeleton and fed into voxel-wise statistics, where
10,000 permutations of the data were generated using randomise
v2.1 with threshold-free cluster enhancement (TFCE) enabled
[26]. The following statistical comparisons were made: (i) crosssectional: very mild (N = 21) and mild (N = 22) Alzheimers disease
patients versus controls (N = 26); and (ii) longitudinal: Alzheimers
disease (N = 16) at 12 months versus baseline data. Note that for
subgroup comparisonsi.e. very mild and mild Alzheimers disease
against controlsand for the longitudinal assessment, we used the
mean FA skeleton derived from all 69 subjects to compute the
Figure 4. Cross-sectional study of mild Alzheimers disease. TBSS results for the mild-stage Alzheimers disease group compared to controls.
Thresholded (TFCE-P,0.05) statistical maps for increased axial/radial diffusivity and reduced FA were overlaid onto the mean FA skeleton and the
MNI152 template. Coronal depths are given in millimetres.
skeletonisation vectors. All statistical maps were thresholded at

a TFCE level of P,0.05 to help prevent the known issue of Type I
errors in voxel-wise experiments.
Regional analysis. Region-of-interest analyses can be confounded in DTI studies where tracts are not directly visualised
such as in the white matter of the cerebral hemispheres. For
instance, the effect of atrophy in a patient group could mean that
regions of interest are subtlythough systematicallymisplaced
with respect to tract anatomy, and this could cause spurious
alterations in tensor metrics. Furthermore, the impact of crossing
fibres can make changes in tensor behaviour difficult to interpret.
For instance, increased l1 has been previously reported in
Alzheimers disease [5,7,8,11], which was a somewhat unanticipated finding as it implied that disease was associated with greater
diffusivity along the preferential orientation of white matter tracts.
An alternate hypothesis to explain this phenomenon, however, was

that it might relate to differential disease involvement in crossing
tracts e.g. if tract A and B cross, but only tract B is affected by
disease, then its degeneration will lead to an apparent increase in
l1and greater anisotropybecause the tensor is now being
influenced more exclusively by tract A [13]. To avoid these
confounds, the midline corpus callosum was studied because (i) it
can be directly visualised with obvious boundary limits and (ii) it is
devoid of crossing fibres. It has further advantages in that it is
a large structure (cf. the fornix, which can also be directly
visualised) hence minimising partial volume effects; finally, prior
knowledge in Alzheimers disease, indicates that one expects
greater degenerative change in the splenium compared to the genu
[7]; hence one can study differential disease effects in the same
white matter bundle.
5

Each T1-weighted structural volume was affine-registered to its

corresponding b0 image, hence enabling the midline corpus
callosum to be individually traced with Analyze v8.1 (Biomedical
Imaging Resource, Mayo Foundation, Rochester, MN, USA),
while minimising partial-volume contamination. Then in Matlab
v2008a (The Mathworks Inc., Natick, MA, USA), we automated
the subdivision of each corpus callosum mask into three regions
splenium, truncus and genu as illustrated in Figure 2 (top panel)of
equal length along the axis that connects the most distalcaudal
and rostralpoints from the mid-sagittal corpus callosums centreof-mass. Note that although rigorous, tractography-based strategies have been proposed to subdivide the corpus callosum [27], in
this study we were ultimately interested in discerning overall tensor
differences between caudal and rostral callosal tracts; thus, for
simplicity, we followed a widely-used previous classification [28]
we collapsed the isthmus and splenium regions into a single
splenial region of interest, and the truncus included both anterior
and posterior sections of the callosal midbody. It should also be
noted that although both schemes were highly concordant in the
caudal third, they disagreed on the definition of genu, which only
extended across the anterior sixth of the corpus callosum in the
most-recently proposed classification [27].
We computed mean values for each region of interest in native
space and for comparison, we also extracted mean valuesdirectly
in standard spaceresulting from the intersection of each region of
interest (segmented from the standard FMRIB58_FA template)
with the mean FA skeleton mask (N = 69, FA.0.2) as depicted in
Figure 2 (bottom panel). It is important to note that the latter
method only required tracing the template corpus callosum mask
once; whereas the former approach needed corpus callosum masks
to be delineated for every individual. Although manual extraction
is considered to be the gold standard method [14], it is a timeconsuming process and in neurodegeneration, atrophy can lead to
systematic misregistration to template of a patient group relative to
controls [29]. In order to circumvent this problem, TBSS projects
tract centres onto a skeleton containing all major white matter
bundles; thereby minimising the effect of misregistration, but also
excluding peripheral white matter tracts. In this study, we also
tested the hypothesis that skeletonised mid-sagittal corpus callosum
DTI data contains all the relevant information needed both to
detect early white matter damage in Alzheimers disease and to
monitor disease progression.
To address the research questions posed in this study, we first
collapsed all DTI measurements from all Alzheimers disease
subjects (N = 43), which, for each callosal subregion, were crosssectionally compared against control data (N = 26) and regressed
against a measure of global cognitive status (ACE-R scores). These
tests aimed to find differential behaviours across regions. Subsequently, the patient group was subdividedas proposed in the
Subjects subsection (very mild, N = 21; and mild Alzheimers
disease, N = 22), and further cross-sectional comparisons were
performed to assess the overall trend of each DTI parameter
across disease severity. The group study designs were as follows:
both very mild and mild patient groups were contrasted against
controls; additionally, mild patients were compared against the
very mild cohort. Further confirmation of the observed tensor
behaviour was sought by testing a smaller group of Alzheimers
patients (N = 16) longitudinally: 12 months versus baseline, and
contrasting data from both time-points against controls.
In Matlab, we first applied a numerical Lilliefors test of the
default null hypothesis that mean-subject values came from
a distribution in the normal family [30], which revealed that they
did not at a = 0.05 for any DTI metric in any cohort. We therefore
compared unpaired DTI-derived values from independent
Figure 5. Cross-sectional results in the mid-sagittal corpus

callosum. TBSS results across the sagittal midline for very mild and
mild Alzheimers disease groups compared to controls.
Table 3. Alzheimers disease skeletonised DTI parametric comparisons in different mid-sagittal callosal areas.
Alzheimers disease (N = 43) vs. Control (N = 26)
Splenium
Truncus
Genu
l1
22.24
21.39
20.22
RD
23.08**
20.76
0.53
MD
23.39**
21.08
0.30
FA
2.451
0.40
21.18
Results are given as Wilcoxon rank-sum Z-statistic.

Significance levels: 10.01,P,0.05 (Alzheimers disease worse than controls, two-tailed); **P,0.05 (Alzheimers disease worse than controls, two-tailed, Bonferronicorrected on n = 12).
samples (i.e. patients versus controls or very mild versus mild

patients) using nonparametric Wilcoxon rank-sumi.e. Mann
Whitney Utests [31,32], and paired longitudinal samples (i.e.
same patient cohort at different time-points) were tested with the
Wilcoxon signed-rank approach [32]. For consistency, all twosample hypotheses on cognitive profiles were tested with the same
relevant (i.e. rank-sum if cross-sectional; signed-rank if longitudinal) method. Linear dependence with ACE-R was tested using
pairwise Pearsons correlations [33]. Note that statistical significances were computed combining both tails of the sampling
distributions.
To assess for atrophy in the midline corpus callosum, we
compared cross-sectional callosal areas in native space. Each
measurement was normalised for differences in total intracranial
volume using an analysis of covariance approach, where the
measured areas were adjusted by an amount proportional to the
difference between each individuals observed total intracranial
volume and the global total intracranial volume mean for all
control subjects [34]. Total brain volumes were also computed and
normalised (by total intracranial volume) to compare (N = 43) midsagittal callosal areas with a measure of global atrophy. Total
intracranial and brain volumes were determined using a previously
validated method that involved summing grey matter, white
matter and cerebrospinal-fluid tissue segments [35].
Simultaneous inference procedure. In order to correct for
the statistical effect of simultaneous regional testing in the present
study, we applied Bonferroni inequalities to each family of
hypotheses with a = 0.05. We corrected inferences separately if
the hypotheses under evaluation were different. For instance, we
performed 28 cross-sectional statistical tests to identify the nature
of cognitive deficits across groups at differing disease stages relative
to controls. This experiment was treated as a family (n = 28 tests);

the family-wise error (FWE) rate associated with a = 0.05 was
therefore PFWE = 0.0018. Analogously, statistical significance
(corrected for multiple comparisons) on longitudinal comparisons
of cognitive performance was established at P,0.0071 (n = 7). The
cross-sectional analyses (N = 43 patients versus controls)performed to characterise the different imaging parameters in each
callosal subdivisionwere treated as a separate family (n = 12,
PFWE = 0.0042). Pearsons correlations were also treated as a family
(n = 12, PFWE = 0.0042). In addition, the cross-sectional assessments of each patient subgroup (i.e. very mild and mild
Alzheimers disease versus controls) were corrected together
(n = 24, PFWE = 0.0021). Note though that for each region of
interest, longitudinal DTI data was corrected separately (n = 12,
PFWE = 0.0042).
Results
Cross-sectional TBSS Study of Very Mild Alzheimers
Disease
TBSS results for the voxel-wise group contrast of very mild
Alzheimers disease versus controls are shown in Figure 3. The l1
statistical map showed significant bilateral and confluent change
that predominantly involved parietal white matter, with strongest
abnormalities along the posterior cingulum and the inter-hemispheric tracts of the caudal corpus callosum. In contrast, there was
relative sparing of caudal occipital, rostral temporal lobe and
prefrontal white matter, and, the more rostral tracts of the corpus
callosum. RD abnormalities spatially overlapped with those found
for l1 in parietal and superior temporal white matterthough only
Table 4. DTI group comparisons across disease stages and linear dependence on global cognition for all patients in the splenium.
Splenium
ACE-R Linear
Dependence (N = 43)
Cross-sectional
Very mild Alzheimers (N = 21) vs.
Control (N = 26)
Mild Alzheimers (N = 22) vs.

Control (N = 26)
Mild Alzheimers (N = 22) vs. Very

mild Alzheimers (N = 21)
l1
2.62*
1.19
1.54
0.15
RD
2.131
3.07**
20.89
20.40**
MD
3.01**
2.74*
0.21
20.21
FA
21.27
22.85*
1.4
0.44**
Group results are given as Wilcoxon rank-sum Z-statistic; statistical dependencies are given as Pearson correlation coefficient with 41 degrees of freedom.
Significance levels: 10.01,P,0.05 (Alzheimers disease worse than controls, two-tailed); *P,0.01 (Alzheimers disease worse than controls, two-tailed); **P,0.05
(Alzheimers disease worse than controls, two-tailed, Bonferroni-corrected on n = 24; or DTI measure correlated with Alzheimers disease cognitive status, two-tailed,
Bonferroni-corrected on n = 12).

Figure 6. Cross-sectional diffusion tensor behaviour in the splenial region. Mean subject values for skeletonised DTI parameters in the
splenium as a function of cognitive status (ACE-R scores) for controls (green), very mild Alzheimers disease (blue) and mild Alzheimers disease
patients (red). The error bars represent 6 one group standard deviation. The vertical axes were scaled to 10 control standard deviations. The vertical
lines delimit the control exclusion criteria (ACE-R,88/100) and the median split (ACE-R = 74). A least-square linear fit was displayed if Pearsons
correlation coefficient was deemed statistically significant (Table 4).
Mid-sagittal Corpus Callosum
across the right hemisphere. FA changes were similarly distributed,

but overall were the least widespread.
TBSS. Unlike total brain volume in Alzheimers disease,

which was significantly reduced compared to controls (standard
score, Z = 20.80; P = 0.002), atrophy in the cross-sectional area of
the corpus callosum did not reach statistical significance
(Z = 20.43, P = 0.1).
Figure 5 shows DTI abnormalities across the mid-sagittal corpus
callosum for both very mild and mild Alzheimers disease stages.
The caudal corpus callosum was heavily involved, whereas there
was preservation of white matter tracts running across the midsagittal genu. l1 featured prominently in the very mild Alzheimers
disease cohort but it did not become more extensive in the moreimpaired group; in contrast, RD/FA abnormalities were more
extensive in mild-stage Alzheimers disease. Note that the fornix
was abnormal for all DTI metrics in both Alzheimers disease
groups (uncorrected-P,0.01, data not shown), but the clusters of
significance did not survive our attempt to correct for multiple
comparisons.
Cross-sectional TBSS Study of Mild Alzheimers Disease

Extensive, mostly bilateral distributions of DTI abnormalities
for all metrics (increased diffusivities and reduced FA) were found
in the mild Alzheimers disease versus control group contrast using
TBSS. As illustrated in Figure 4, l1 showed the most extensive and
confluent clusters of significance. These were located in parietal
white matter regions including the caudal corpus callosum and the
posterior cingulum bundle; caudal temporal areas with a slightly
greater predilection for the left side and bilateral frontal lobe
involvement Although RD and FA abnormalities were highly
concordant and largely overlapped with l1 clusters, overall they
were less extensive; in particular, frontal lobe involvement of RD
and FA was minimal. Note that all regions found to be abnormal
in the very mild Alzheimers disease group (Figure 3), were also
clearly damaged in the mild group (Figure 4).
The cross-sectional MD results for both Alzheimers disease
groups were also extensive and unsurprisingly overlapped to a high
degree with those of its components; more specifically with the
spatial distribution of increased l1, but it did not reveal any
additional involvement (see Figure S1).
Regions of Interest
The regional analyses were consistent with the above TBSS
observations. Table 3 highlights the preservation of the midline
truncus and genu relative to the splenium when contrasting all
Alzheimers patients (N = 43) with the control group. Examining
8
Figure 7. Longitudinal study of Alzheimers disease. Whole-brain TBSS contrast on 12-month follow-up versus baseline in the longitudinal
Alzheimers disease cohort (TFCE-P,0.05) for increased radial diffusivity and reduced FA n.b. no significant results for l1 or MD were found.
10
for differential behaviour of the diffusion tensor across varying

disease stages reinforced the finding that in early disease, l1 is
more prominently abnormal than RD indicating that a significant
contributor to early MD increase is an increase in l1 (Table 4).
RD and FA, however, showed the most prominent alterations in
the more cognitively-impaired group of patientswhere l1
remained relatively stable, suggesting that MD abnormalities in
later disease stages are primarily driven by increased RD (Table 4).
The latter finding was supported by a statistically significant
correlation between dementia severity (ACE-R score) and either
FA or RD (Table 4). It should be noted that statistical tests for the
truncus and genuequivalent to those reported in Table 4 for the
spleniumyielded, as expected, no significant results.
Focusing on the splenial lesion, this being the key pathological
region, Figure 6 illustrates all mean patient DTI values relative to
control data as a function of cognitive status. As shown in Table 4,

splenial l1 appeared to increase significantly in very mild subjects,
but it did not follow that more impaired patients were increasingly
abnormalif anything there was a trend to attenuation of
abnormality with advancing dementia (Figure 6a). In stark
contrast, RD and FA were essentially less sensitive in very mild
disease but there was a linear progression observed for these
behaviours with advancing severity (Figure 6b and 6d). Note that
the behaviour of MD followed a similar pattern to that of l1 in
early disease stages but appeared to be largely driven by RD in
more advanced cases (Figure 6c).
It should be noted that the regional results reported in this
sectionusing skeletonised DTI datawere replicated with data
extracted from entire callosal regions of interest in native space (see
plots in Figure S2).

nal pairs (Figure 9b and 9d, respectively) followed more coherently

progressive evolutions than axial and mean diffusivities (Figure 9a
and 9c).
Discussion
This study aimed to investigate the biomarker potential of DTI
metrics in Alzheimers disease through whole-brain analyses as
well as by taking a reductionist approach in the corpus callosum to
eliminate ambiguities generated by uncertain tract visualisation
and crossing fibres. The evolution of DTI changes was examined
in both cross-sectional and longitudinal datasets; the former by
way of a median spilt in which 76% of the very mild patients were
in the mild-cognitive impairment stage of Alzheimers disease, and
the latter by contrasting 12-month follow-up scans to baseline.
Across the various analyses, a consistent picture emerged in which
an increase in l1 is the first significant change in Alzheimers
disease, but then remains relatively steady. In contrast, RDand
therefore FAbecome progressively more abnormal with disease
evolution, making them the candidate metrics for stage-specific
biomarkers. In other words, these markers could have value to
track change over time. Given that FA reduction is a function of
both RD and l1, and the early increase in l1 appeared to
attenuate with disease progression, it may be that FA is the
superior measure of decline compared to RD in longitudinal
studies. It should be emphasized, however, that the attenuation of
l1 was subtle and, unless replicated in similar studies, could simply
represent random variability in the current dataset. A further
caveat to this proposal is that the preference for FA would only
apply to areas in which the l1 abnormality has already peaked
such as the splenium in the present analyses. For instance, in
a repeated-measure longitudinal design, in areas that are initially
spared in terms of l1, the first movement of this metric would be to
increase, leading to a loss of sensitivity for FA to detect change.
This was exemplified by the cross-sectional results, in which frontal
and left temporal white matter was normal in the very mild group,
but became abnormal (mostly in terms of l1) in the mild group.
This problem would, of course, be avoided by studying RD,
whichunlike FAis independent of l1.
In contrast to RD and FA, the results indicated that l1 has no
role at all as staging biomarker for a given region of interest.
Although one could, in theory, propose it as a staging marker by
looking at the differential spatial extent of l1 abnormalities over
time, this would be difficult to implement as a marker of
progression given that some degree of heterogeneity would be
expected across subjects; furthermore, such an approach would
also be confounded if it proved correct in future studies that the l1
Figure 8. Longitudinal results in the mid-sagittal corpus

callosum. Longitudinal TBSS results for radial diffusivity and fractional
anisotropy across the midline.
Longitudinal Study of Alzheimers Disease

The contrast of 12-month follow-up scans with baseline in the
longitudinal Alzheimers disease cohort (N = 16) found that l1
remained unchanged in the whole brains skeletonised white
matter at TFCE-P,0.05. However, consistent with the crosssectional analyses, RD/FA abnormalities progressed significantly
in the caudal corpus callosum and posterior cingulum bilaterally,
and in a left superior temporal region (Figure 7). The clusters of
significance found in the longitudinal assessments were highly colocalised with the strongest DTI abnormalities found in the very
mild Alzheimers disease group relative to controls; this was also
clearly illustrated in the mid-sagittal TBSS results (Figure 8).
The splenial results shown in Table 5 revealed a very consistent
scenario. l1 was slightly abnormal at baseline but it did not
progress. RD and FA differences, however, above the statistical
threshold at baselineshowed emphatic abnormalities at 12
months and during the longitudinal time-span. MD was found
to be similarly abnormal at baseline and 12 months, thus resulting
in no apparent progression.
When the extracted mean DTI values from each subjects
skeletonised splenium were plotted against ACE-R scores
(Figure 9), the findings were in remarkable agreement with those
observed in the cross-sectional study. Overall, RD/FA longitudi-
Table 5. Longitudinal DTI assessment of Alzheimers disease in the splenium.
Longitudinal
Splenium
Baseline (N = 16) vs.
Control (N = 26)
12 months (N = 16) vs.

Control (N = 26)
12 months (N = 16) vs.

Baseline (N = 16)
l1
2.371
1.54
20.52
RD
1.72
2.97**
22.59*
2.71*
21.76
22.58*
22.64*
MD
2.47
FA
20.87
Baseline and 12 months versus controls comparisons are reported as Wilcoxon rank-sum Z-statistic values; paired longitudinal results are given as Wilcoxon signed-rank
Z-statistic.
Significance levels: 10.01,P,0.05 (Alzheimers disease worse than controls, two-tailed); *P,0.01 (Alzheimers disease worse than controls; or 12 months worse than
baseline, two-tailed); **P,0.05 (Alzheimers disease worse than controls, two-tailed, Bonferroni-corrected on n = 12).
10
11
Figure 9. Longitudinal tensor behaviour in the splenium. Longitudinal pairs of mean subject skeletonised DTI parameters as a function of
cognitive status (ACE-R) for Alzheimers disease subjects at baseline (blue) and 12 months (red).
12
were increased l1 and MD whereas, also consistent with our

findings, in areas such as the cingulum and fornix that would be
expected to have most advanced pathology, RD and FA were
highly abnormal but l1 was not.
It was recently proposed that l1 increase might represent
differential tract involvement in areas of crossing white matter
tracts [13]i.e. if two tracts cross then the tensor in such areas will
be the average contribution of both; therefore, if one of these tracts
degenerates, the tensor will be driven by the remaining tract and
hence an increase in l1 could be predicted. While this seems a very
plausible theory in crossing fibres, it cannot explain the l1 increase
in very early Alzheimers disease because the observation is equally
apparent in areas such as the splenium [5,7] and fornix [5,37], in
which crossing tracts are not present. One clue from the current
study was the finding that increased l1 in the splenium showed no
correlation with global dementia severity. This, in turn, implies
that increased l1 might be completely independent of axonal loss.
It must be stressed, nevertheless, that the failure of l1 to correlate
with dementia severity is only demonstrated at a global level. It
may still be the case that the phenomenon driving the early
increase in l1 is associated with local dysfunction of specific
cognitive processes. For instance, a recent study noted correlation
between increased l1 and route-learning impairment in Alzheimers disease [38]; the area of correlation was tightly co-incident
with correlations between the same task and both metabolism and
effect attenuates in early affected regions. Nevertheless, the results

of both cross-sectional and longitudinal studies clearly indicate that
an increase in l1 is the first sign of change in Alzheimers disease,
suggesting therefore that it could have a role as an early statespecific marker. This was particular notable in the cross-sectional
data, in which posterior temporo-parietal white matter including
the splenium showed increased l1 in the very mild group with the
same regions showing involvement of RD and FA in the mild
group i.e. at a more advanced disease stage. In the more advanced
group, however, new areasnotably frontal and left temporal white
mattershowed increased l1, suggesting that as degeneration
spreads to new areas, an increase in l1 is the first sign of this
involvement. Similarly, the longitudinal data also showed that
there was a prominent increase in l1 in posterior temporo-parietal
white matter at baseline (cross-sectional comparison against the
control population not shown), with RD and FA abnormalities
emerging in the same spatial distribution with follow-up.
Increased l1 as found in this and other recent studies [5,7] was
an unanticipated finding in Alzheimers disease and its mechanism
is uncertain. The finding that the l1 abnormalities were the most
spatially extensive extends the results of our earlier study using the
TBSS method [5], and are also consistent with a recent
independent TBSS study in showing that this metric has the
greatest sensitivity to change in early disease [36]. Using a similar
TBSS-based approach, Huang et al. [7] also found that the most
prominent features in less affected areas in Alzheimers disease
11

atrophy of adjacent grey matter suggesting that it was not an

artefactual result.
Studies in Alzheimers disease, to date, have mostly tended to
conceptualise DTI abnormalities in terms of neuronal loss or
damage, but it is important to keep in mind that although robust
tensor changes can be found in Alzheimers disease, their precise
underlying mechanism is essentially unknown. Furthermore,
studies often infer mechanisms by citing homology with animal
models, but this can only ever be consistent with rather than proof of
a given mechanism; for instance, if myelin loss in an animal model
causes a particular tensor metric to change in a certain way, it does
not follow that myelin loss is the only mechanism that could
generate such a tensor behaviour. Returning to the present
findings, RD, and therefore FA, correlated with dementia severity
implying that whatever these metrics capture, it is likely to be
related to neuronal loss. The fact that increased l1 was both
independent of dementia severity and a precursor to RD/FA
changes (n.b. areas where RD and FA changes emerged were
characteristically those that had first seen an increase in l1),
suggests that l1 may be capturing an upstream event to axonal
degeneration. It is also noteworthy that the spread of l1 increase
was not random but, rather, closely mirrored the expected
progression of degeneration as would be measured by cerebral
glucose metabolismi.e. posterior (posterior cingulate/precuneus
spreading out to lateral temporo-parietal association tracts), then
later to anterior (prefrontal) areas [39,40].
Although a completely speculative hypothesis, one possible
upstream event could be inflammation. The role of inflammatory
change in Alzheimers disease is presently of considerable interest
[41] and it is at least conceivable, given that l1 increase appears to
represent something other than axonal loss, that it could be driven
by factors such as microglial activation. Whatever its mechanism,
understanding the true sequence in the cascade of events leading
to neuronal loss in Alzheimers disease is arguably the most
important question in understanding pathogenesis. As such,
further work to understand what this early l1 increase might
represent in areas that subsequently degenerate, seems to be,
therefore, critically important.
The differing behaviour of tensor metrics identified in the
course of Alzheimers disease in this study goes some way to
explain inconsistencies reported in previous studies. For instance,
some studies in early disease have reported little abnormality in FA
[5,6]. That said, it is difficult to identify a coherent picture from
previous DTI results in Alzheimers disease, even if one only
considers the corpus callosumfor review see [42]. Factors
contributing to this inconsistency likely include: differing levels of
dementia severity; an almost universal lack of outcome data in
cohorts labelled mild-cognitive impairment; variable region-ofinterest placement; and variable acquisition protocols. In the
present study, the cross-sectional and longitudinal data highlighted
the critical effect of dementia severity, while clinical follow-up
ensured that all cases scanned in the mild-cognitive impairment
stage had incipient Alzheimers disease. The issue of region-ofinterest placement is critical to standardisation if DTI data is to be
used as a biomarker. Because TBSS fits tract centres common to
all subjects, this appears to be an appropriate method for wholebrain analyses. Meanwhile, the midline corpus callosum offers
a cleaner target to study a tract in isolation; critically, however, as
would be expected from prior knowledge, it is not homogenously
affected but, rather, the splenium is the site of maximum damage.
The current study offers an automated method for extracting
corpus callosum regions that could go some way to optimising
comparability across studies. Finally, the issue of acquisition is
critical to consider. The present study used a 63-direction
acquisition with one b-value, and as already mentioned, yielded

a spread of abnormalities with advancing severity that is highly
consistent with the evolution of Alzheimers disease from other
imaging modalities. Recent methods work has validated the use of
30 directions with two b-values as an alternative, whereas fewer
(than 30) directions may be less reliable for tensor modelling unless
a large number of b-values are used [43]; to this end, it is
important to note that most early DTI studies employed less than
10 directions with only one b-value, which may have contributed
to noisy or inconsistent results. It should also be noted that at
present, although RD, MD and FA are considered to be reliable
DTI metrics, conflicting results have been reported with regard to
the intra-scanner stability of l1 measurements at 3T [44,45]. In
this study, l1 appeared to be robust in the corpus callosum, but
further test-retest validations using the proposed experimental
methodology are required to address this important question. A
final technical point to note is that although it is discouraged to
interpret primary diffusivities on the basis of the underlying
neurobiological processes [46], the negligible effects of: (i) crossing
fibres, (ii) partial volumedue to lack of atrophyand (iii)
eigenvector sorting biasdue to the large differences in signal-tonoise ratio between l1 and RD e.g. l1 < 6?RD in the skeletonised
splenium, makes the mid-sagittal corpus callosum a suitable white
matter structure to monitor early-stage Alzheimers disease.
In conclusion, the cross-sectional and longitudinal data examined in the present study identified that RD and FA have utility as
staging biomarkers in Alzheimers disease. Both whole-brain
TBSS and a skeletonised splenial region of interest appear to be
good methods to standardise sampling in repeated-measures
designs. The sensitivity of these measures compared to other
biomarkers needs to be established, however, one final point
argues for their potential added value in practical terms. Structural
MRI is already a standard procedure in longitudinal studies;
a diffusion sequence can be readily acquired in the same scanning
session, meaning that this added value can be achieved with little
extra cost or inconvenience compared to nuclear medicine
imaging or cerebrospinal-fluid analysis. Axial diffusivity (l1) in
a given region is a state-, rather than a stage-specific biomarker,
that predates changes in RD/FA and appears to represent
something other than axonal loss.
Supporting Information
Figure S1 Increased mean diffusivity results for very
mild and mild Alzheimers disease. TBSS results for very
mild- and mild-stage Alzheimers disease groups compared to
controls. Thresholded (TFCE-P,0.05) statistical maps for increased mean diffusivity were overlaid onto the mean FA skeleton
and the MNI152 template with coronal depths given in
millimetres. Extensive, mostly bilateral distributions of DTI
abnormalities for increased mean diffusion were found in the
very mild and mild Alzheimers disease group comparisons.
Significant abnormalities were located in parietal white matter
regions including the caudal corpus callosum and the posterior
cingulum bundle, and in caudal temporal areas. All clusters of
significance found in the very mild Alzheimers disease group,
were also found in the mild group. As expected, overall MD
abnormalities were highly concordant and largely overlapped with
the spatial distribution of l1 clusters of significance shown in
Figure 3 and Figure 4 (main manuscript).
(TIF)
Figure S2 Regional analysis of the splenium in native
space. Mean subject values for DTI parameters in the native
splenial region as a function of cognitive status (ACE-R) for
12
13
controls (green), very mild Alzheimers disease (blue) and mild

Alzheimers disease patients (red). The error bars represent 6 one
group standard deviation, and the vertical axes were scaled to 10
control standard deviations. The vertical lines delimit the control
exclusion criteria (ACE-R,88) and the median split (ACER = 74). A least-square linear fit was displayed if Pearsons
correlation coefficient was deemed statistically significant for
N = 43 patients. In agreement with the skeletonised data shown
in Figure 6 (main manuscript), l1 appeared to increase most
significantly in very mild subjects, whereas RD and FA followed
a linear progression with advancing disease severity.
(TIF)
Acknowledgments
We are grateful for the support from our patients and healthy volunteers.
Author Contributions
Conceived and designed the experiments: JAC GBW PJN. Performed the
experiments: JAC GP. Analyzed the data: JAC SA. Wrote the paper: JAC
GBW GP PJN.
References
14
1. Jack CR, Jr., Petersen RC, Xu YC, OBrien PC, Smith GE, et al. (1999)
Prediction of AD with MRI-based hippocampal volume in mild cognitive
impairment. Neurology 52: 13971403.
2. Fellgiebel A, Schermuly I, Gerhard A, Keller I, Albrecht J, et al. (2008)
Functional relevant loss of long association fibre tracts integrity in early
Alzheimers disease. Neuropsychologia 46: 16981706.
3. Mielke MM, Kozauer NA, Chan KC, George M, Toroney J, et al. (2009)
Regionally-specific diffusion tensor imaging in mild cognitive impairment and
Alzheimers disease. Neuroimage 46: 4755.
4. Takahashi S, Yonezawa H, Takahashi J, Kudo M, Inoue T, et al. (2002)
Selective reduction of diffusion anisotropy in white matter of Alzheimer disease
brains measured by 3.0 Tesla magnetic resonance imaging. Neurosci Lett 332:
4548.
5. Acosta-Cabronero J, Williams GB, Pengas G, Nestor PJ (2010) Absolute
diffusivities define the landscape of white matter degeneration in Alzheimers
disease. Brain 133: 529539.
6. Fellgiebel A, Wille P, Muller MJ, Winterer G, Scheurich A, et al. (2004)
Ultrastructural hippocampal and white matter alterations in mild cognitive
impairment: a diffusion tensor imaging study. Dement Geriatr Cogn Disord 18:
101108.
7. Huang H, Fan X, Weiner M, Martin-Cook K, Xiao G, et al. (2012) Distinctive
disruption patterns of white matter tracts in Alzheimers disease with full
diffusion tensor characterization. Neurobiol Aging 33: 20292045.
8. Shu N, Wang Z, Qi Z, Li K, He Y (2011) Multiple diffusion indices reveals white
matter degeneration in Alzheimers disease and mild cognitive impairment:
a tract-based spatial statistics study. J Alzheimers Dis 26 Suppl 3: 275285.
9. Zhang Y, Schuff N, Du AT, Rosen HJ, Kramer JH, et al. (2009) White matter
damage in frontotemporal dementia and Alzheimers disease measured by
diffusion MRI. Brain 132: 25792592.
10. Huang J, Auchus AP (2007) Diffusion tensor imaging of normal appearing white
matter and its correlation with cognitive functioning in mild cognitive
impairment and Alzheimers disease. Ann N Y Acad Sci 1097: 259264.
11. Bosch B, Arenaza-Urquijo EM, Rami L, Sala-Llonch R, Junque C, et al. (2012)
Multiple DTI index analysis in normal aging, amnestic MCI and AD.
Relationship with neuropsychological performance. Neurobiol Aging 33: 6174.
12. Zhang Y, Schuff N, Jahng GH, Bayne W, Mori S, et al. (2007) Diffusion tensor
imaging of cingulum fibers in mild cognitive impairment and Alzheimer disease.
Neurology 68: 1319.
13. Douaud G, Jbabdi S, Behrens TE, Menke RA, Gass A, et al. (2011) DTI
measures in crossing-fibre areas: increased diffusion anisotropy reveals early
white matter alteration in MCI and mild Alzheimers disease. Neuroimage 55:
880890.
14. Fellgiebel A, Muller MJ, Wille P, Dellani PR, Scheurich A, et al. (2005) Colorcoded diffusion-tensor-imaging of posterior cingulate fiber tracts in mild
cognitive impairment. Neurobiol Aging 26: 11931198.
15. Zhang J, Jones M, DeBoy CA, Reich DS, Farrell JA, et al. (2009) Diffusion
tensor magnetic resonance imaging of Wallerian degeneration in rat spinal cord
after dorsal root axotomy. J Neurosci 29: 31603171.
16. Hanyu H, Asano T, Sakurai H, Imon Y, Iwamoto T, et al. (1999) Diffusionweighted and magnetization transfer imaging of the corpus callosum in
Alzheimers disease. J Neurol Sci 167: 3744.
17. McKhann G, Drachman D, Folstein M, Katzman R, Price D, et al. (1984)
Clinical diagnosis of Alzheimers disease: report of the NINCDS-ADRDA Work
Group under the auspices of Department of Health and Human Services Task
Force on Alzheimers Disease. Neurology 34: 939944.
18. Folstein MF, Folstein SE, McHugh PR (1975) "Mini-mental state". A practical
method for grading the cognitive state of patients for the clinician. J Psychiatr
Res 12: 189198.
19. Mioshi E, Dawson K, Mitchell J, Arnold R, Hodges JR (2006) The
Addenbrookes Cognitive Examination Revised (ACE-R): a brief cognitive test
battery for dementia screening. Int J Geriatr Psychiatry 21: 10781085.
20. Reese TG, Heid O, Weisskoff RM, Wedeen VJ (2003) Reduction of eddycurrent-induced distortion in diffusion MRI using a twice-refocused spin echo.
Magn Reson Med 49: 177182.
21. Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, et al. (2002)
Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn
Reson Med 47: 12021210.
22. Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, Behrens TE, et al.
(2004) Advances in functional and structural MR image analysis and
implementation as FSL. Neuroimage 23 Suppl 1: S208219.
23. Jenkinson M, Smith S (2001) A global optimisation method for robust affine
registration of brain images. Med Image Anal 5: 143156.
24. Smith SM (2002) Fast robust automated brain extraction. Hum Brain Mapp 17:
143155.
25. Smith SM, Jenkinson M, Johansen-Berg H, Rueckert D, Nichols TE, et al.
(2006) Tract-based spatial statistics: voxelwise analysis of multi-subject diffusion
data. Neuroimage 31: 14871505.
26. Smith SM, Nichols TE (2009) Threshold-free cluster enhancement: addressing
problems of smoothing, threshold dependence and localisation in cluster
inference. Neuroimage 44: 8398.
27. Hofer S, Frahm J (2006) Topography of the human corpus callosum revisited
comprehensive fiber tractography using diffusion tensor magnetic resonance
imaging. Neuroimage 32: 989994.
28. Witelson SF (1989) Hand and sex differences in the isthmus and genu of the
human corpus callosum. A postmortem morphological study. Brain 112: 799
835.
29. Snook L, Plewes C, Beaulieu C (2007) Voxel based versus region of interest
analysis in diffusion tensor imaging of neurodevelopment. Neuroimage 34: 243
252.
30. Lilliefors HW (1967) On the Kolmogorov-Smirnov test for normality with mean
and variance unknown. Journal of the American Statistical Association 62: 399
402.
31. Mann HB, Whitney DR (1947) On a Test of Whether one of Two Random
Variables is Stochastically Larger than the Other. Annals of Mathematical
Statistics 18: 5060.
32. Wilcoxon F (1946) Individual comparisons of grouped data by ranking methods.
J Econ Entomol 39: 269.
33. Pearson K (1896) Mathematical Contributions to the Theory of Evolution. III.
Regression, Heredity and Panmixia. Philosophical Transactions of the Royal
Society A 187: 253318.
34. Jack CR, Jr., Twomey CK, Zinsmeister AR, Sharbrough FW, Petersen RC, et
al. (1989) Anterior temporal lobes and hippocampal formations: normative
volumetric measurements from MR images in young adults. Radiology 172:
549554.
35. Pengas G, Pereira JM, Williams GB, Nestor PJ (2009) Comparative reliability of
total intracranial volume estimation methods and the influence of atrophy in
a longitudinal semantic dementia cohort. J Neuroimaging 19: 3746.
36. ODwyer L, Lamberton F, Bokde AL, Ewers M, Faluyi YO, et al. (2011)
Multiple indices of diffusion identifies white matter damage in mild cognitive
impairment and Alzheimers disease. PLoS One 6: e21745.
37. Oishi K, Akhter K, Mielke M, Ceritoglu C, Zhang J, et al. (2011) Multi-modal
MRI analysis with disease-specific spatial filtering: initial testing to predict mild
cognitive impairment patients who convert to Alzheimers disease. Front Neurol
2: 54.
38. Pengas G, Williams GB, Acosta-Cabronero J, Ash TWJ, Hong YT, et al. (2012)
The relationship of topographical memory performance to regional neurodegeneration in Alzheimers disease. Front Ag Neurosci 4: 17.
39. Chase TN, Foster NL, Fedio P, Brooks R, Mansi L, et al. (1984) Regional
cortical dysfunction in Alzheimers disease as determined by positron emission
tomography. Ann Neurol 15 Suppl: S170174.
40. Nestor PJ, Fryer TD, Smielewski P, Hodges JR (2003) Limbic hypometabolism
in Alzheimers disease and mild cognitive impairment. Ann Neurol 54: 343351.
41. Wyss-Coray T, Rogers J (2012) Inflammation in Alzheimer disease-a brief
review of the basic science and clinical literature. Cold Spring Harb Perspect
Med 2: a006346.
42. Di Paola M, Spalletta G, Caltagirone C (2010) In vivo structural neuroanatomy
of corpus callosum in Alzheimers disease and mild cognitive impairment using
different MRI techniques: a review. J Alzheimers Dis 20: 6795.
13

43. Correia MM, Carpenter TA, Williams GB (2009) Looking for the optimal DTI
acquisition scheme given a maximum scan time: are more b-values a waste of
time? Magn Reson Imaging 27: 163175.
44. Danielian LE, Iwata NK, Thomasson DM, Floeter MK (2010) Reliability of
fiber tracking measurements in diffusion tensor imaging for longitudinal study.
Neuroimage 49: 15721580.
45. Takao H, Hayashi N, Kabasawa H, Ohtomo K (2012) Effect of scanner in

longitudinal diffusion tensor imaging studies. Hum Brain Mapp 33: 466477.
46. Wheeler-Kingshott CA, Cercignani M (2009) About "axial" and "radial"
diffusivities. Magn Reson Med 61: 12551260.
15
14
Cost-Effectiveness of Magnetic Resonance Imaging with

a New Contrast Agent for the Early Diagnosis of
Alzheimers Disease
Maria Biasutti1,2, Natacha Dufour1,2, Clotilde Ferroud3, William Dab1, Laura Temime1*
1 Laboratoire Modelisation et Surveillance des risques sanitaires Conservatoire national des Arts et Metiers 75003, Paris, France, 2 Ecole des Ponts ParisTech 77700,
Marne la Vallee, France, 3 Laboratoire Transformations Chimiques et Pharmaceutiques Conservatoire national des Arts et Metiers 75003, Paris, France
Abstract
Background: Used as contrast agents for brain magnetic resonance imaging (MRI), markers for beta-amyloid deposits might
allow early diagnosis of Alzheimers disease (AD). We evaluated the cost-effectiveness of such a diagnostic test, MRI+CLP
(contrastophore-linker-pharmacophore), should it become clinically available.
Methodology/Principal Findings: We compared the cost-effectiveness of MRI+CLP to that of standard diagnosis using
currently available cognition tests and of standard MRI, and investigated the impact of a hypothetical treatment efficient in
early AD. The primary analysis was based on the current French context for 70-year-old patients with Mild Cognitive
Impairment (MCI). In alternative screen and treat scenarios, we analyzed the consequences of systematic screenings of
over-60 individuals (either population-wide or restricted to the ApoE4 genotype population). We used a Markov model of
AD progression; model parameters, as well as incurred costs and quality-of-life weights in France were taken from the
literature. We performed univariate and probabilistic multivariate sensitivity analyses. The base-case preferred strategy was
the standard MRI diagnosis strategy. In the primary analysis however, MRI+CLP could become the preferred strategy under a
wide array of scenarios involving lower cost and/or higher sensitivity or specificity. By contrast, in the screen and treat
analyses, the probability of MRI+CLP becoming the preferred strategy remained lower than 5%.
Conclusions/Significance: It is thought that anti-beta-amyloid compounds might halt the development of dementia in
early stage patients. This study suggests that, even should such treatments become available, systematically screening the
over-60 population for AD would only become cost-effective with highly specific tests able to diagnose early stages of the
disease. However, offering a new diagnostic test based on beta-amyloid markers to elderly patients with MCI might prove
cost-effective.
Citation: Biasutti M, Dufour N, Ferroud C, Dab W, Temime L (2012) Cost-Effectiveness of Magnetic Resonance Imaging with a New Contrast Agent for the Early
Diagnosis of Alzheimers Disease. PLoS ONE 7(4): e35559. doi:10.1371/journal.pone.0035559
Editor: David W. Dowdy, Johns Hopkins Bloomberg School of Public Health, United States of America
Received July 22, 2011; Accepted March 20, 2012; Published April 20, 2012
Copyright: 2012 Biasutti et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This research was financed by a grant from the Cnam, a public higher education establishment by which all authors were employed at the time. The
funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
produced, how it accumulates and how it is eliminated. Animal

studies in mice have suggested that anti-beta-amyloid drugs can
reduce brain amyloid level and improve memory problems in
diseases similar to AD. At the present time, there is no clear
evidence that these drugs can improve Alzheimer symptoms or
protect brain cells but it is thought that they could halt the
development of dementia in patients with early stage AD [5].
With these prospects of further therapeutic developments,
attention has now focused on improving the sensitivity and
specificity of diagnostic tools, and in developing tools that would
allow early diagnosis.
The current standard diagnostic strategy of AD generally
comprises a detailed history, a standardized assessment of
cognition and functional status and laboratory testing. Brain
imaging examinations such as nonenhanced computed tomography imaging, positron emission tomography imaging, or magnetic
Introduction
Alzheimers disease (AD) is the main cause of dementia in older
people, with approximately 26 million cases worldwide [1,2,3].
What is more, a major increase in this prevalence is expected in
years to come [3]. In France, while 850.000 people were
diagnosed with AD in 2004, 2.1 million may be affected by
2040 [4].
There are currently no treatments that may cure AD or halt the
course of the disease. However, in recent years, drugs such as
acetylcholinesterase inhibitors have demonstrated efficacy at
reducing the intensity of certain symptoms. Moreover, new
avenues for research are being investigated. Many scientists
believe that one of the main causes of the AD has to do with betaamyloid, microscopic protein which accumulate throughout the
cortex of Alzheimer patients. This is called the amyloid
hypothesis [2]. They believe that the destruction of brain cells
seen in AD is caused by defects in the way beta-amyloid is
16
PLoS ONE | www.plosone.org
April 2012 | Volume 7 | Issue 4 | e35559

Cost-Effectiveness of MRI+CLP for Alzheimers
1.1. Primary analysis. In the primary analysis, three

diagnostic strategies were compared over a three year period for
a cohort of 70 year-old individuals consulting for the first time
following mild cognitive impairment (MCI) symptoms:
resonance imaging (MRI) are also sometimes used in order to

exclude other conditions or to measure brain atrophy.
Finding more accurate diagnostic tools implies discovering noninvasive sensitive and specific biomarkers for AD. One avenue of
research lies in the detection of b-amyloid plaques [6,7]. This
detection could be achieved through the use of new contrast agents
for MRI which bind to b-amyloid plaques, thus allowing a valid
diagnosis of AD at a very early stage [6,8].
A few studies have attempted to assess the cost-effectiveness of
imaging diagnosis tools for AD [9,10,11], or to evaluate the impact
of screening the general population [12]. However, none has
focused on combinations of MRI and new contrast agents.
In this study, we evaluate the cost-effectiveness of a diagnostic
strategy based on MRI with a new contrast agent detecting bamyloid plaques (contrastophore-linker-pharmacophore or CLP).
We compare this strategy with other current standard diagnostic
strategies. In the primary analysis, the new strategy is simply
introduced as an alternative to current diagnostic tools and made
available to the same population. In this setting, we investigate the
consequences of the introduction of a new AD treatment with
significant efficacy at an early stage of the disease.
In alternative scenarios, we assume that the availability of this
new treatment would naturally raise the issue of the opportunity of
screening for AD. We thus evaluate the cost-effectiveness of
diagnostic strategies in the context of hypothetical national
screening programs.
the standard diagnosis strategy based on an interview with an

AD specialist, cognition tests such as mini-mental state
evaluation (MMSE) and laboratory tests (standard diagnosis
strategy)
a strategy combining the standard strategy with standard MRI
(standard MRI strategy)
and a strategy combining the standard strategy with MRI used
with a new contrast agent detecting b-amyloid plaques
(MRI+CLP strategy)
Figure 1A depicts the tree showing the possible outcomes of
using one of the investigated diagnostic tests in this scenario. The
choice of a 3-year horizon for this analysis was motivated by data
on the duration of efficacy of currently available AD treatments
and mortality rates of over-70 dementia patients.
As part of the sensitivity analysis, we assumed that a new
treatment had been developed which delayed significantly the
course of AD at an early stage and that this treatment was offered
to MCI patients diagnosed with AD but with high MMSE scores
(who were either false positives or early AD patients).
1.2. Screen and treat analyses. In the screen and
treat analyses, it was assumed that a new treatment had been
developed which delayed significantly the course of AD at an early
stage, and that a national screening campaign targeted at
individuals over 60 years-old took place. Screening could either
be population-wide or limited to individuals carrying the e4 allele
of the apolipoprotein E gene (ApoE4), in whom the risk of AD has
been shown to be significantly higher [16].
In this context, three diagnostic strategies were compared over a
fifteen year period for a cohort of 60 year-old individuals taking
part in the screening campaign:
Methods
1. Framework of the cost-effectiveness analyses
We performed cost-effectiveness analyses in order to compare
the costs and benefits of several alternative diagnostic strategies in
the French context. Costs were measured in Euros (J) and benefits
were measured in quality-adjusted life-years (QALYs), which
assign to each year of life a weight between 0 (dead) and 1 (perfect
health).
A strategy was considered dominated if another strategy had a
better or similar efficacy at a lower cost; conversely, a strategy was
considered strongly dominant when it was both more effective and
cheaper than all other strategies. We computed incremental costeffectiveness ratios (ICERs), in which changes in resource use,
compared with the next best strategy, were included in the
numerator, while additional health effects, compared with the next
best strategy, were included in the denominator. Finally, we
compared ICERs to the willingness-to-pay (WTP) for an
additional QALY, which was assumed equal to three times the
gross national product per capita in France, as recommended by
the World Health Organization Choice working group [13,14],
that is, 76.171J per QALY in 2009. If no strategy was strongly
dominant, the preferred strategy was that with the highest ICER
under the willingness-to-pay threshold.
The study was conducted from a societal perspective, meaning it
included all costs and benefits, no matter who incurred them.
Future costs and QALYs were discounted at 5% annually.
The cost-effectiveness analyses were conducted using the
TreeAge software [15]. We performed expected value analyses,
based on the computation over simulated cycles of percentages of
a hypothetical cohort in each modeled AD stage. For the
multivariate sensitivity analyses, we also performed Monte Carlo
simulations with 10.000 trials, in order to derive the distribution of
incremental cost-effectiveness ratios for the MRI+CLP strategies,
as well as acceptability curves for all strategies.
Table 1 details the compared strategies.
the standard diagnosis strategy,

the standard MRI,
and the MRI+CLP strategy
Figure 1B depicts the tree showing the possible outcomes of the
screening procedure with one of the investigated diagnostic tests in
this scenario.
Following the initial screening, individuals who were not
diagnosed as AD patients were screened again every 5 years,
unless they developed dementia symptoms between these scheduled screenings, in which case they were immediately tested.
The choice of a 15 year horizon for this second analysis was
motivated by the need to follow 60 year-old mostly healthy
individuals through several screening campaigns.
2. Models of disease progression

We used a Markov model of the evolution of AD based on
previous work [17] in which the disease evolved in 5 stages: no
AD, mild AD, moderate AD, severe AD and death. For the screen
and treat analyses, a 6th stage was added for asymptomatic patients
with early AD. We also distinguished patients that were taken care
of at home from institutionalized patients, at each stage of AD.
Initially, all patients in the cohort began in the community setting
and had a probability in each cycle, conditional on disease state
(and, for the 2d analysis, on age), of making a transition to nursing
home care. File S1 and Figure S1 present the models we used in
more detail.
2
17
Table 1. Compared strategies.
Strategy
Tests
Imaging
Standard diagnosis
Laboratory tests, clinical examination, cognition test
None
Standard MRI
Non-enhanced MRI
MRI+CLP
MRI with a new contrast agent detecting b-amyloid plaques
Considering the natural history of AD, we chose to model

disease progression using 6 month cycles. Transition probabilities
over 6 months were obtained from the square root of the matrix of
annual transition probabilities which were based on the literature
18
[17,18,19] and available data on age-specific mortality rates [20]

and age-specific AD incidence [4] in France.
Table S1 lists all model parameters, with their base-case values.
Figure 1. Decision tree for an individual tested for Alzheimers disease (AD). Possible outcomes of the testing procedure are depicted as a
function of the individuals health status for: (a) The primary scenario (testing of over-70 patients consulting for dementia symptoms). (b) The screen
and treat scenario (systematic screening of the over-60 population). Depending on the investigated strategy, the generic diagnostic test
mentioned in the trees may be standard diagnosis, standard MRI or MRI+CLP. When AD is diagnosed, the imaging procedures are followed by a
cognition test (MMSE) in order to determine the disease stage. No test is performed in severe AD patients, who are assumed to be diagnosed directly.

In the second screen and treat analysis, screening was

restricted to individuals carrying the e4 allele of the apolipoprotein
E gene (ApoE4). Based on earlier studies showing significantly
higher risk of AD occurrence in ApoE4 individuals [16], we
assumed that age-specific incidence rates were doubled.
3. Available treatments
The model assumes that all patients who receive a diagnosis of
probable Alzheimer disease receive treatment with donepezil,
memantine or a hypothetical higher-efficacy drug.
If the patient is diagnosed at a mild or moderate stage,
donepezil is prescribed, to be replaced by memantine when the
patient has evolved to a severe stage of AD. If the patient is
diagnosed when the disease is already severe, memantine is
prescribed.
In addition, a hypothetical treatment T which is efficient at
early stages of AD may be prescribed to patients who are
diagnosed with AD but have high MMSE scores in the primary
analysis, or to patients who are diagnosed with early AD in the
screen and treat analyses.
Donepezil treatment has been shown to improve significantly
memory and other cognitive functions in patients with mild to
moderate AD [21], and to reduce the annual decline in cognition
in these patients when compared with patients in a placebo group
[22]; similarly, memantine treatment has been shown to cause a
clinically noticeable reduction in deterioration over 28 weeks,
compared with placebo, in patients with moderate-to-severe AD
[23,24]. It is to be noted that actually, donepezil or memantine
help treat the symptoms of AD although there is no evidence that
they modify the underlying pathology of the disease. On the basis
of these data, we assumed that treatment modified transition
probabilities between disease stages, both reducing the speed of
AD progression and increasing the chance of symptom lessening
(Table S1). For treatment T, we only assumed a decrease in the
progression from early to mild and moderate stage AD, as we
supposed that reversal to asymptomatic AD of mild stage patients
was not possible. In the base-case analysis, these effects were
assumed to be constant throughout the duration of treatment;
long-term clinical studies suggest that treatment efficacy may last
for up to 3 years [21,25].
5. Diagnostic tests
The standard diagnostic strategy of AD was assumed to
comprise a detailed history of the patient, an assessment of
cognition and functional status using a questionnaire test such as
the MMSE, and laboratory testing. Other strategies combined
MRI (with or without a new contrast agent) for AD diagnosis with
a questionnaire test such as the MMSE aimed at determining the
stage of the disease.
In the case of severe AD, it was assumed that dementia
symptoms allowed direct diagnosis over a consultation (without
need for a diagnostic test). Hence, all severe stage AD patients
were assumed to be diagnosed and to receive treatment.
As in [10], we estimated the sensitivity of the standard
diagnostic tests at 75% and their specificity at 90%; in early stage
AD, we assumed that patients were asymptomatic and that the
standard diagnostic therefore performed as it did in non-AD
patients, declaring only 10% ( = 120.9) of them as having AD.
Based on data from a clinical study, we also hypothesized a
sensitivity of 50% in early stage, 88% in mild stage and 95% in
moderate stage AD, as well as a 96% specificity, for standard MRI
diagnosis [31].
Regarding the hypothetical diagnostic test using MRI with the
new contrast agent (MRI+CLP), we based our assumptions on
available data on PET-scan amyloid imaging. As studies on
amyloid plaques suggest that amyloid deposition reaches a plateau
by the early clinical stages of Alzheimers disease (amyloid cascade
hypothesis), we assumed the sensitivity of the MRI+CLP
diagnostic test to be independent of the disease stage. In one
study on PET tracers, the sensitivity and specificity were estimated
at 90% [32]; in another, a 95% sensitivity and 83% specificity
were found [33]. Here, we assumed a 96% sensitivity and 87%
specificity for MRI+CLP.
4. AD prevalence and incidence

In the primary analysis, the simulated cohort comprised 70 year
old individuals who consulted for cognitive impairment symptoms.
There is evidence that older persons with Mild Cognitive
Impairment (MCI) feature neurobiological AD in 50% to 70%
of the cases [26]. Based on this evidence and on data from a study
at Massachusetts General Hospital [10], we estimated that 56% of
the cohort patients had AD initially. The distribution between
disease stages at this first consultation was estimated, based on a
French study [27], at 55.9% mild stage AD, 39.9% moderate stage
AD and 4.2% severe stage AD.
MCI is often a precursor to Alzheimers dementia and the
annual rate of development of AD for patients with MCI is 10 to
15% [26,28]. Here, we assumed a 10%/year AD incidence in
untreated patients of the cohort.
In the first screen and treat analysis, the cohort was
representative from the French population of individuals over 60
years old, as it was assumed that a national screening was
underway. Therefore the initial prevalence of AD (including early
asymptomatic stage AD) in the cohort was assumed to be 1% [3].
We further assumed that there was an 8/7 ratio between the
prevalences of early stage and mild stage AD [Personal
communication, Pr. Verny, AP-HP, Paris, 2009]. The distribution
between mild, moderate and severe stages was the same as in the
primary analysis.
Age-specific AD incidence rates for individuals between 60 and
75 years old were chosen to be consistent with data from recent
cohort studies [29,30].
6. Costs and effectiveness

All costs were measured at their 2009 level.
The cost of the standard
diagnosis was computed as the sum of the cost of a specialist
consultation in France, estimated at 55J [34], that of mental state
evaluation tests, estimated at 69J [35], and that of standard
laboratory tests, estimated at 50J [36]. The cost of MRI was
obtained from the Classification Commune des Actes
Medicaux, a fixed-costs scale of medical procedures based on
practitioners fees, fixed costs for the medical procedures
themselves, and fixed costs for operating the equipment [35].
Finally, we estimated the cost of the new contrast product for MRI
at 250J [Personal Communication, Guerbet company, Paris,
2009].
6.2. Costs of AD follow-up. Diagnosed AD patients were
assumed to have a follow-up consultation every 6 month with an
AD specialist. These specialist consultations were estimated at 41J
[34].
6.3. Costs of treatment. We used prices for generic AD
drugs, that is, 572J per 6-month period for generic donepezil
(Aricept) and 286J per 6-month period for generic memantine
(Ebixa) [37]. We further assumed that the hypothetical new drug
that would be efficient in early stage AD would have a cost similar
to that of generic donepezil.
6.1. Costs of diagnostic tests.
19
20
6.4. Costs of care. We took into account both living and care
costs. For institutionalized patients, living costs included hostel
costs, based on a French study of AD patients [18], and costs of
care included caregiver wages. In an earlier study, caregivers were
estimated to spend 517 hours over a 3 month period caring for
moderate-to-severe AD patients [38]. Here, we therefore
computed costs associated with caregivers over 6 months by
multiplying hourly wages (estimated at 13 J/h for professional
caregivers) with 1034 hours.
For patients living at home, basic costs included living and
medical expenditures [39], based on a French study of home-cared
AD patients and controls; the cost of care by unprofessional
caregivers was also estimated as an opportunity cost. In order to do
this, we valued informal care at a conservative value of 8.4 J/h,
and we assumed that these unpaid caregivers spent the same
amount of time caring for AD patients than professional
caregivers, that is, 1034 hours over a 6 month cycle.
6.5. Indirect costs. In this study, we took into account
several indirect costs of AD. First, we assigned an opportunity cost
to unprofessional caregivers who took care of AD patients leaving
at home, as described before.
Second, we took into account the burden of AD care on the
health state of unprofessional caregivers. Based on an earlier study
[38], 35% of AD caregivers take medication related to their
activity; they have high rates of depression and anxiety, as well as
high overall morbidity and mortality rates, compared to non-AD
caregiver controls [40]. We estimated the cost of this health impact
as that as of a weekly psychiatrist consultation, plus that of an
antidepressant treatment, for 35% of unprofessional caregivers.
Finally, we also evaluated the cost associated with the loss of
productivity of AD patients. This is generally not done in costeffectiveness studies of AD, since AD patients are for the most part
retired. However, recent data shows that pensioners are becoming
more and more involved in volunteer activities within nonprofit
organizations (NPOs), as well as performing informal volunteer
activities [41].
Here, based on French data [42], we estimated that 60 to 75
year-old individuals performed on average 63.8 hours of informal
volunteering activities over a 6 month period mostly within the
family sphere, such as childcare for instance. This was valued at
7.7J/hour, which was the minimum wage in France in 2009, and
multiplied by an efficiency coefficient of 0.7 to arbitrarily take
into account the reduced productivity in older individuals.
Similarly, we estimated that 50% of 60 to 75 year-old
individuals are involved in NPOs, with a mean of 12 hours of
volunteer activities per month. On average, we hence estimated
that 60 to 75 year-old individuals performed a total of 36 hours of
volunteer activities within NPOs over a 6 month period [43],
which were valued at 7.9J/hour and multiplied by the
aforementioned 0.7 efficiency coefficient.
We added the resulting estimated productivity benefit of a
French pensioner to our analyses as a cost associated to AD, in full
for moderate-to-severe patients and multiplied by 0.6 for mild AD
patients (assuming that mild AD only reduces productivity by
40%).
6.6. Effectiveness. We estimated quality-of-life weights
(QALYs) for over-60 patients without Alzheimer disease at 0.826
on a scale of 0 to 1, on the basis of the mean of time trade-off
scores for men and women aged 6584 years old published in a
study of health outcomes in the general population [44].
Quality-of-life weights for patients with Alzheimer disease at
each disease stage and care setting (institution or community) were
based on previously published Health Utilities Index Mark 2
(HUI:2) scores [17].
7. Sensitivity analysis
The ranges investigated in the Sensitivity Analysis are
summarized in Table S1, along with the data sources. We
performed both univariate and multivariate sensitivity analyses.
For the multivariate analyses, we performed a Monte-Carlo
simulation with 10.000 trials, using a priori triangular distributions
for model parameters. We then identified the most influential
parameters in the cost-effectiveness of the MRI+CLP strategy by
calculating the partial rank correlation coefficient (PRCC) between
each input parameter and the ICER of this strategy and assessing
their statistical significance.
7.1. Drug effects and prices. We investigated the impact of
the introduction of new drugs, which would have the same costs
and indications as donepezil and memantine, respectively, but
with varying efficacy: the probabilities of transitions under
treatment from mild to moderate or moderate to severe stage
were further reduced by a factor (fmM or fMS) ranging from 0.5 to
1, and the probabilities of transitions under treatment from
moderate to mild or severe to moderate stage were further
increased by a factor (fMm or fSM) ranging from 1 to 2 [17]. For
simplicity reasons, we also summarized these four avenues for
improvement of current AD treatment through a single parameter
f, assuming a linear relationship between the multiplying factors:
f~f ML ~f SM ~{2|f LM z3
Regarding the hypothetical drug efficient for early stage AD
(treatment T), we investigated both more and less efficient drugs,
with probabilities of progression from early to mild and moderate
stage AD under treatment T ranging from 0 to their values
without treatment T; fT was the reduction factor applied to these
transition probabilities due to treatment T (in [01]).
The assumed cost for a 6-month cure with treatment T was also
varied, between 0 and 1000J.
7.2. Diagnostic tests characteristics. We investigated
sensitivities from 0.9 to 1 and specificities from 0.7 to 1 in early
to moderate stage AD patients for the hypothetical diagnostic test
using MRI with the CLP contrast product. We also investigated
sensitivities ranging from 0.75 to 0.90 for the standard diagnosis in
moderate stage AD.
In the screen and treat analyses, we investigated sensitivities
of standard MRI between 0.1 and 0.5 for early stage AD.
7.3. Disease progression. To model faster or slower
progression of AD, we investigated state transition probabilities
in our Markov model ranging from 10% lower to 10% higher than
their base-case values.
7.4. Initial distribution of patients and prevalence. In
the primary analysis, we investigated AD prevalences among
consulting individuals ranging from 50 to 70%. For each fixed
prevalence, we varied the proportion of patients with mild AD
between 50 and 75% of all AD patients; the ratio between
moderate and severe AD prevalence remained the same, that is,
10 moderate stage patients for 1 severe stage patient.
In the screen and treat analyses, we investigated initial AD
prevalences among screened individuals ranging from 1 to 10%.
For each fixed prevalence, we varied the proportion of patients
with early asymptomatic AD from 30 to 75% of all AD patients;
the distribution of mild to severe stages remained the same.
7.5. Discount rate. As the screen and treat analyses
spanned a 15 year period, we assessed the impact of variations in
the assumed discount rate for costs and QALYs, from 0 to 10%
per year.
7.6. Frequency of screenings. In the screen and treat
analyses, we investigated the impact of varying durations between
screening campaigns, ranging from one to ten years.
5

T). Assuming either more effective AD treatments, higher speed of

AD progression, a larger initial AD prevalence in 70 year-old MCI
individuals, a larger initial portion of mild AD, higher sensitivity or
specificity of MRI+CLP diagnosis or a lower cost of the CLP
product could lead to MRI+CLP becoming the preferred strategy,
with an ICER lower than the willingness-to-pay threshold.
Table 3 (columns 1 and 2) provides partial rank correlation
coefficients (PRCC) between input values and the ICER of the
MRI+CLP strategy (compared with the standard MRI strategy,
without treatment T). Only parameters whose PRCC with this
ICER is statistically significant at the confidence level of 5% or
lower are shown, namely: the speed of disease progression; the
initial AD prevalence in the studied population; the initial portion
of mild stage AD; the discount rate; the sensitivity and specificity of
the MRI+CLP diagnostic test; and the cost of the CLP contrast
agent.
In Figure 2, the sensitivity of our conclusions on the MRI+CLP
diagnostic test characteristics (sensitivity, specificity and cost) is
assessed. MRI+CLP is the preferred strategy in a wide array of
scenarios, including some assuming lower sensitivity or specificity
or higher cost than in the base-case.
Figure S2 provides the distribution of incremental costeffectiveness ratios for the MRI+CLP strategy, compared with
standard diagnosis, as well as an acceptability curve depicting the
probability for each of the three investigated strategies to be
preferred as a function of the willingness-to-pay threshold. For
willingness-to-pay thresholds larger than 90 000J/QALY,
MRI+CLP becomes the preferred strategy.
With the assumed WTP threshold at 76 171J/QALY, the
probability that MRI+CLP was the preferred strategy increased
from 43% to 64% when treatment T was assumed to be available
(data not shown). Figure S3 investigates the impact of the efficacy
and cost of treatment T on these results.
Table 2. Results of the primary analysis (base-case

hypothesis): computed cost, efficacy and cost-effectiveness
(C/E) ratio of the standard diagnosis, standard MRI and
MRI+CLP strategies, and incremental cost-effectiveness ratios
(ICER) of the MRI+CLP strategy as compared with the standard
MRI strategy.
Efficacy (in
QALYs)
C/E
ICER
Standard diagnosis 36 294
1.7663
20 548
Dominated
Standard MRI
36 131
1.7710
20 401
MRI+CLP
36 313
1.7731
20 480
Strategy
Cost (in J)
In the current context
88 439 J/QALY
Assuming hypothetical treatment T has been made available

1.7668
20 523
Standard MRI
36 117
1.7712
20 391
MRI+CLP
36 268
1.7737
20 447
Dominated
60 923 J/QALY
Results
1. Primary analysis
1.1. Base Case. The first part of Table 2 summarizes the
results of the primary cost-effectiveness analysis in the base case; in
the second part of Table 2, hypothetical treatment T is offered to
all MCI patients diagnosed with AD but with high MMSE scores.
In both cases, the standard diagnosis strategy was dominated by
the standard MRI strategy (more costly and less effective).
Cost and effectiveness increased from standard MRI to
MRI+CLP strategies. In the base-case, the ICER of the
MRI+CLP strategy was higher than the French willingness-topay threshold (estimated at 76 171 J/QALY). Hence, standard
MRI was found to be the preferred strategy. However, assuming
that treatment T had been made available led to a lower ICER for
MRI+CLP, making it the preferred strategy.
1.2. Sensitivity analysis. Detailed results of the univariate
sensitivity analysis are presented in Table S2 (without treatment
2. Screen and treat analyses

2.1. Base Case. Table 4 summarizes the results of the
screen and treat cost-effectiveness analyses in the base case; in
the first part, the whole over-60 population is screened, while in
the second part, the screening is targeted at ApoE4 individuals. In
both cases, the standard diagnosis strategy was dominated by the
standard MRI strategy.
Table 3. Partial rank correlation coefficients (PRCC) between input values and the ICER of the MRI+CLP strategy (compared with
the preferred strategy).
Primary cost-effectiveness analysis
Screen and treat cost-effectiveness analysis (population-wide screening)
Parameter
PRCC
Parameter
PRCC
Speed of disease progression
20,41
Initial prevalence of AD
20,86
Initial prevalence of AD
20,51
Initial portion of early AD
20,15
Initial portion of mild AD
20,26
Sensitivity of MRI+CLP test
20,17
Sensitivity of MRI+CLP test
20,92
Specificity of MRI+CLP test
20,64
Specificity of MRI+CLP test
20,73
Discount rate
0,29
Discount rate
0,19
Cost of the CLP contrast agent
0,41
Cost of the CLP contrast agent
0,79
Cost of treatment T
0,78
Impact of treatment T (from high to low)
0,41
Speed of disease progression
20,23
Sensitivity of standard MRI in early AD
0,51
All PRCCs are statistically significant at the confidence level of 5%. A higher absolute value of PRCC indicates a strong relationship between that parameter and the ICER;
a positive (resp. negative) value of PRCC implies that the value of the ICER increases (resp. decreases) when the value of the input increases.
21
22

Figure 2. Results of the primary analysis: multivariate sensitivity analysis. The strategy with maximum net monetary benefit is depicted as
a function of the assumed sensitivity and specificity of the MRI+CLP diagnostic test, for assumed costs of the CLP contrast agent between 0 and 500J
per injection (in the absence of treatment T): (a) Cost of the CLP contrast agent at 50 J/injection. (b) Cost of the CLP contrast agent at 250 J/
injection. (c) Cost of the CLP contrast agent over 450 J/injection.
Although overall computed costs and QALYs were lower when

assuming a screening targeted at the ApoE4 population, the main
base-case results were similar in both our screening analyses. The
MRI+CLP strategy was both more costly and more efficient than
the standard MRI strategy. Its ICER, compared with the standard
MRI strategy, was much higher than the French willingness-to-pay
estimated at 76 171 J/QALY. Therefore, standard MRI was the
preferred strategy in the base case.
2.2. Sensitivity analysis. Detailed results of the univariate
sensitivity analysis are presented in Table S3 for the populationwide screening scenario. For an assumed specificity of MRI+CLP
higher than 98%, MRI+CLP became the preferred strategy; it was
strongly dominant for specificities over 99%. No other individual
model parameter had enough impact on our results to make
MRI+CLP the preferred strategy.
Table 3 (columns 3 and 4) provides partial rank correlation
coefficients (PRCC) between inputs values and the ICER of the
MRI+CLP strategy (compared with the standard MRI strategy),
for the population-wide screening scenario. Only parameters
whose PRCC with this ICER is statistically significant at the
confidence level of 5% or lower are shown, namely: the initial AD
prevalence in the studied population; the specificity of the
MRI+CLP diagnostic test; and the efficacy and cost of new
treatment T.
Figure 3 depicts results from a multivariate analysis performed
on the efficacy and cost of the hypothetical new treatment and
MRI+CLP specificity, while Figure 4 depicts results from a
multivariate analysis performed on the initial prevalence of AD in
the general over-60 population, the specificity of MRI+CLP
diagnosis and the assumed cost of the CLP contrast agent. Several
combinations of these parameters allow for MRI+CLP strategy
dominance, but all include high specificity of MRI+CLP diagnosis.
For instance:
half, MRI+CLP is the preferred strategy for the base-case cost

of treatment T (500J/6 months of treatment)
assuming an initial AD prevalence of 4% and the base-case cost
of 250J per injection for the CLP contrast agent, MRI+CLP is
the preferred strategy as long as MRI+CLP diagnosis is more
than 97% specific.
Figure S4 provides the distribution of incremental costeffectiveness ratios for the MRI+CLP strategy, compared with
standard MRI, as well as an acceptability curve depicting the
probability for the MRI+CLP strategy to be the preferred strategy
as a function of the willingness-to-pay threshold. This probability
remains lower than 4% even assuming a willingness-to-pay at
200 000J/QALY.
When assuming a screening targeted at the ApoE4 population,
the probability that MRI+CLP was the preferred strategy
increased slightly, reaching 5% for a 200 000 J/QALY
willingness-to-pay threshold (data not shown).
Discussion
Here, we investigated the cost-effectiveness of a new diagnostic
tool for AD allowing early diagnosis in two different contexts. First,
we assumed that this new diagnostic tool would be made available
to the same individuals who are currently offered other diagnostic
tests. Second, we hypothesized that a treatment with significant
efficacy in early stage AD was developed, and that, as a
consequence, a national screening campaign was put into place,
using either the currently available diagnostic tests or the new
diagnostic test.
We found that in both analyses, the preferred base-case strategy
was the standard MRI strategy. However, multivariate sensitivity
analyses showed that, while in the primary analysis combining
MRI with a new contrast product could prove the preferred
strategy under a wide array of realistic scenarios, the probability of
this happening was inferior to 5% in the hypothesis of a national
screening campaign. Even assuming the availability of a low-cost
highly efficient treatment in early AD, novel contrast agents would
need to have very high specificity to be cost-effective when used for
systematic screening of the entire population.
assuming 98% specificity for MRI+CLP diagnosis and that

treatment T reduces AD progression from the early stage by
Table 4. Results of the screen and treat analyses (base-case
hypothesis): computed cost, efficacy and cost-effectiveness
(C/E) ratio of the standard diagnosis, standard MRI and
MRI+CLP strategies, and incremental cost-effectiveness ratio
(ICER) of the MRI+CLP strategy as compared with the standard
MRI strategy.
Strategy
Efficacy (in
Cost (in J) QALYs)
C/E
ICER
Dominated
1. Models of disease progression

We used Markov models of AD progression. Although widely
used in similar cost-effectiveness analyses [10,11,17,18,45,46], this
approach presents several limitations. First, it has been shown that
using time-independent transition probabilities may lead to either
overestimate or underestimate disease progression, depending on
the AD stage [47]. Second, in order for the model complexity to
remain manageable, we were led to limit the number of states. In
order to take into account indirect consequences of AD, such as
the loss of productivity of AD patients in their volunteer activities,
we incorporated them as aggregate factors in the computation of
costs.
Using other modeling approaches, such as discrete events
simulation, would allow a more detailed and realistic description of
AD progression and, in turn, derive more reliable conclusions
[12,48].
Population-wide screening
8.0722
5 396
Standard MRI
43 009
8.0732
5 327
MRI+CLP
44 945
8.0752
5 566
991 972 J/QALY
Dominated
Screening targeted at ApoE4 individuals

8.0377
5 563
Standard MRI
44 180
8.0386
5 496
MRI+CLP
46 075
8.0415
5 730
641 326 J/QALY
23
24

Figure 3. Results of the screen and treat (population-wide screening) analysis: multivariate sensitivity analysis. The strategy with
maximum net monetary benefit is depicted as a function of the assumed efficacy and cost of the hypothetical new drug T, for assumed specificities of
the MRI+CLP diagnosis test between 0.80 and 0.99. The efficacy of treatment T is expressed as a 0-to-1 ratio between assumed probabilities of
transition from early stage AD with and without treatment T; 0 corresponds to maximum efficacy and 1 to no efficacy (in the base case, fT = 0.5: 50%
reduction). Only costs lower than the base-case cost of 500J per 6-month treatment are investigated here. (a) specificity for MRI+CLP inferior to 0.97
(including base case). (b) 0.98 specificity for MRI+CLP. (c) 0.99 specificity for MRI+CLP.
2. Assumed properties of the MRI+CLP diagnostic test
standard MRI, we also explored how our results changed when

assuming higher sensitivity and specificity for the standard
diagnosis, or lower sensitivity and specificity for standard MRI.
Irrespective of the assumed specificity, the standard diagnosis
strategy remained dominated by the standard MRI strategy.
However, as shown in Figure S6, standard diagnosis could become
preferable to standard MRI under a wide array of combinations of
lower sensitivity for standard MRI and higher sensitivity for
standard diagnosis in moderate AD.
All in all, it should be underlined here that comparing standard
MRI to the current standard diagnosis strategy for AD was not the
focus of this study and that our results should not be taken as
evidence that large-scale screening using standard MRI should be
undertaken.
Several limitations of this analysis lie with the assumptions we
made on AD treatment.
First, we neglected all adverse effects from the treatments.
Including a cost associated with such adverse effects in treated
individuals would lead to reduced ICERs for all strategies. As costs
would be expected to increase while QALYs remained unchanged,
the ICER of the MRI+CLP strategy should only increase.
Second, we assumed that all diagnosed individuals were treated.
This is in reality probably false, as AD patients may end up
rejecting treatment or not taking it properly.
Finally, another limitation of this analysis is that we did not take
into account the psychological impact of early diagnosis of AD in
the absence of efficient treatment at this stage. Including this factor
would increase the cost associated to the MRI+CLP strategy, and
in turn its ICER.
Because the new CLP contrast agent for MRI is still under
development, assumptions had to be made regarding its sensitivity
and specificity for diagnosing AD. As mentioned in the Methods,
we based these assumptions in part on available data on PET-scan
used with amyloid markers [32,33]. We also investigated relatively
wide ranges for these characteristics in our sensitivity analyses.
Regarding the specificity, it should be noted that amyloid
plaques have been detected in 10 to 30% of otherwise apparently
normal elderly subjects. This may limit the capacity of MRI+CLP
to correctly identify healthy subjects, which is why we investigated
specificities as low as 0.70 in our sensitivity analyses. However,
several studies also suggest that the presence of amyloid plaques is
associated with declining cognitive test scores and with structural
and functional brain changes suggestive of early AD [49]. In two
recent follow-up studies, about one third of patients with mild
cognitive impairment in whom amyloid plaques were detected
converted to AD over the following 2 years [50,51]. It could
therefore be argued that at least part of these 1030% amyloidpositive individuals may not be false positive, but very early
stage AD patients.
Figure S5 illustrates the influence of false positives and false
negatives associated with the MRI+CLP diagnosis in terms of
additional cost per QALY in both our primary analysis and our
screen and treat analysis.
3. Included costs
In this study, we chose to include indirect costs in addition to
direct costs. The importance of these indirect costs, which include
costs associated with the loss of productivity of AD patients and
costs associated with informal caregiving at the homes of AD
patients, has indeed been underlined in several recent studies,
which showed that they may dominate direct costs of care in early
stages of the disease [52]. As a consequence, it has been suggested
in recent reviews that future cost-effectiveness studies should take
into account such indirect costs [48,53].
5. Results and limits of the screen and treat analyses

In our screen and treat analyses, we investigated a
hypothetical future in which new treatments effective in early
stage AD were available. To our knowledge, ours is the first study
to do so, although earlier works investigated improvements in
currently available treatments for mild to severe stage AD [10,11].
Innovatively, this second analysis also included a community-wide
screening program and repeated rounds of testing. Indeed, if such
treatments were to become available, population screening
programs which are not currently recommended by any health
Agency would have to be discussed.
Because the investigated context was hypothetical, we had to
make several assumptions which influenced our results. First, we
had to estimate the prevalence of AD in the general over-60
population, including asymptomatic cases. There is obviously no
data to document this prevalence, and we chose a rather
conservative value of 1%; the actual figure may be well over this
value and we investigated prevalences up to 10% in our sensitivity
analysis. Our base-case analysis was a worst-case scenario for the
MRI+CLP strategy, which included the only diagnosis tool
adapted to early AD.
Second, as both the new treatment and the new diagnosis test
were hypothetical, we chose values for their base-case characteristics based on personal communications with AD specialists and
teams currently working on the development of CLP tracers
[Personal Communication, Guerbet company, Paris, 2009], and
4. Results and limits of the primary analysis

The results we obtained in our primary analysis are similar to
those from previously published cost-efficacy studies, in the sense
that in the base-case, adding a new marker to MRI was not costeffective [10,11]. However, in our multivariate sensitivity analysis,
there was a high probability of MRI+CLP becoming the preferred
strategy.
In addition, it is to be noted that we found the standard MRI
strategy to dominate the standard diagnosis strategy, whereas
earlier studies found very high ICERs for similar imaging
strategies [10,11]. This may be due to the fact that in these
earlier studies, MRI was supposed to come as an addition to the
standard diagnosis procedure, which included a detailed history, a
questionnaire test such as the MMSE, laboratory testing and
nonenhanced CT imaging. In our study, the standard MRI
strategy only comprised the initial specialist consultation, the
MMSE and MRI itself.
Because of the high level of uncertainty surrounding the
assumed sensitivity and specificity of both standard diagnosis and
10
25
26
11

Figure 4. Results of the screen and treat (population-wide screening) analysis: multivariate sensitivity analysis. The strategy with
maximum net monetary benefit is depicted as a function of the assumed prevalence of AD in the general over-60 population (between 0 and 10%)
and specificity of the MRI+CLP diagnostic test between 0.90 and 1, for assumed costs of the CLP contrast agent between 0 and 500J per injection: (a)
Cost of the CLP contrast agent at 50 J/injection. (b) Cost of the CLP contrast agent at 250 J/injection (base-case). (c) Cost of the CLP contrast agent
at 500 J/injection.
investigated wide ranges of variation for these values in our

sensitivity analysis. It is to be noted that the new treatment T may
well be more expensive than 1000J, the top limit of our
investigated range; however, as shown by our results, MRI+CLP
would then systematically not prove cost-effective, irrespective of
other characteristics.
Third, the participation rate to the screening campaign was
unknown, and assumed to be 100%. Based on observed data from
similar screening campaigns, actual rates are more around 60% at
best, meaning that the screened population may not be
representative of the general over-60 population. Also, being
screened positive for AD may have a psychological impact,
especially in asymptomatic patients. We chose not to take into
account this impact in our analysis, considering that it should be
reduced by the availability of a treatment efficient from early AD.
Future studies should model screening participation and impact
more realistically and investigate the potential repercussions.
Finally, this second analysis suffered from the same limitations
related to AD treatment as the first. Including adverse effects of
treatments in this analysis would increase the cost of the
MRI+CLP strategy. Moreover, as the time horizon for this
analysis was 15 years, it would have been interesting to investigate
the impact of varying durations of treatment efficacy. This will be
made easier in future years when more data on the long-term
impact of AD treatment becomes available.
Because there is evidence that individuals carrying the ApoE4
gene are at increased risk for AD, we felt it was pertinent to
investigate a scenario under which screening was targeted at these
individuals. It should be noted however, that, in order to fully
investigate such a scenario, other factors may have to be taken into
account. Indeed, some epidemiological studies suggest that the
ApoE4 genotype increases mortality rates, in particular cardiovascular mortality in older individuals [54], hence age-specific
mortality rates would need to be obtained for ApoE4 individuals.
In addition, amyloid plaque density in cognitively healthy adults
has been shown to be higher in ApoE4 carriers [55]. Depending
on the interpretation of this finding, this may have different
implications for a diagnostic tool based on the detection of amyloid
plaques. On the one hand, it could be assumed that ApoE4
carriage increases the speed of amyloid plaque accumulation in
AD patients; in that case, MRI+CLP would have better sensitivity
in early stage AD in ApoE4 carriers. On the other hand, it could
also be assumed that the ApoE4 genotype increases the density of
amyloid plaques irrespective of AD; in that case, MRI+CLP would
be expected to have lower specificity in ApoE4 carriers.
When this research yields results, a new cost-effectiveness

analysis should be performed in order to evaluate the available
tools with observed data.
Markov models of Alzheimers disease progression. All states are further subdivided in two, for individuals
living at home vs. inside an institution (retirement or nursing
home). A) Model for the primary analysis. B) Model for the
screen and treat analyses.
(TIF)
Figure S1
Figure S2 Multivariate sensitivity analysis of the primary cost-effectiveness study. A) Distribution of incremental
cost-effectiveness ratios (ICER) of the MRI+CLP strategy,
compared to the standard diagnosis strategy. B) Acceptability
curve: probability that each strategy either is dominant or has an
ICER inferior to the willingness-to-pay, as a function of the
willingness-to-pay threshold.
(TIF)
Figure S3 Multivariate sensitivity analysis of the primary cost-effectiveness study with treatment T. The
strategy with maximum net monetary benefit is depicted as a
function of the assumed efficacy and cost of the hypothetical new
drug T. The efficacy of treatment T is expressed as a 0-to-1 ratio
between assumed probabilities of transition from early stage AD
with and without treatment T; 0 corresponds to maximum efficacy
and 1 to no efficacy (in the base case, fT = 0.5: 50% reduction).
(TIF)
Figure S4 Multivariate sensitivity analysis of the
screen and treat (population-wide screening) costeffectiveness study. A) Distribution of incremental costeffectiveness ratios (ICER) of the MRI+CLP strategy, compared
to the standard MRI strategy. B) Acceptability curve: probability
that the MRI+CLP strategy either is dominant or has an ICER
inferior to the willingness-to-pay, as a function of the willingnessto-pay threshold.
(TIF)
Analysis of the influence of MRI+CLP false

positives and false negatives on the cost-effectiveness of
the MRI+CLP strategies (tornado diagram). The range of
incremental cost-effectiveness ratios (ICER) of the MRI+CLP
strategy (compared to the standard MRI strategy) in A) the
primary analysis and B) the screen and treat analysis is depicted
for MRI+CLP sensitivities ranging from 0.9 to 1 and specificities
ranging from 0.7 to 1. A sensitivity lower than 0.96 (base-case
value) implies a higher risk of false negatives (FN, dotted green
bars) than in the base-case. A specificity lower than 0.87 (base-case
value) implies a higher risk of false positives (FP, dotted yellow
bars) than in the base-case.
(TIF)
Figure S5
6. Conclusions
Assuming that a treatment with proven efficacy in early AD
becomes available, as well as a diagnostic test allowing early
detection of the disease, the issue of screening the population will
arise. Our study suggests that, in order for this screening to be costeffective, key parameters are the specificity of the new diagnostic
test and the cost and effectiveness of the new treatment. These
preliminary results ought to be taken into account in the currently
underway research on early detection and treatment of AD,
including work on b-amyloid plaques detection and elimination.
Multivariate sensitivity analysis of the results

of the primary cost-effectiveness study on standard MRI
vs. standard diagnosis. The strategy with maximum net
monetary benefit among these two strategies is depicted as a
Figure S6
12
27
compared to the reference strategy, in the primary costeffectiveness analysis, depending on the values of model
parameters.
(DOCX)
function of the assumed sensitivity of the standard MRI diagnostic

test in mild and moderate AD, for an assumed sensitivity of the
standard diagnosis strategy in moderate AD between 0.75 and
0.85 (in the absence of CLP and treatment T). A) 0.75 sensitivity in
moderate AD for standard diagnosis (base-case value). B) 0.85
sensitivity in moderate AD for standard diagnosis.
(TIF)
Univariate sensitivity analysis: ICER of the

MRI+CLP strategy compared with the reference strategy, in the universal screen and treat cost-effectiveness
analysis, depending on the values of model parameters.
(DOCX)
Table S3
File S1 Models of disease progression.
(DOCX)
Table S1 Model parameters: base-case values and
ranges investigated in the sensitivity analyses.

(DOCX)
Conceived and designed the experiments: CF WD LT. Performed the

experiments: MB ND LT. Analyzed the data: MB ND LT. Wrote the
paper: WD LT. Helped formulate realistic assumptions: CF WD.
Univariate sensitivity analysis: incremental

cost-effectiveness ratio (ICER) of the MRI+CLP strategy,
Table S2
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
28
25. Atri A, Shaughnessy LW, Locascio JJ, Growdon JH (2008) Long-term course
and effectiveness of combination therapy in Alzheimer disease. Alzheimer Dis
Assoc Disord 22: 209221.
26. Drago V, Babiloni C, Bartres-Faz D, Caroli A, Bosch B, et al. (2011) Disease
tracking markers for Alzheimers disease at the prodromal (MCI) stage.
J Alzheimers Dis 26 Suppl 3: 159199.
27. Blum-Boisgard C, Ulmann P (2007) Medicaments specifiques de la maladie
dAlzheimer. Regime Social des Independants. pp 1130.
28. Bowen J, Teri L, Kukull W, McCormick W, McCurry SM, et al. (1997)
Progression to dementia in patients with isolated memory loss. Lancet 349:
763765.
29. Commenges D, Joly P, Letenneur L, Dartigues JF (2004) Incidence and
mortality of Alzheimers disease or dementia using an illness-death model. Stat
Med 23: 199210.
30. Katz MJ, Lipton RB, Hall CB, Zimmerman ME, Sanders AE, et al. (2011) Agespecific and Sex-specific Prevalence and Incidence of Mild Cognitive
Impairment, Dementia, and Alzheimer Dementia in Blacks and Whites: A
Report From the Einstein Aging Study. Alzheimer Dis Assoc Disord.
31. Harris GJ, Lewis RF, Satlin A, English CD, Scott TM, et al. (1998) Dynamic
susceptibility contrast MR imaging of regional cerebral blood volume in
Alzheimer disease: a promising alternative to nuclear medicine. AJNR
Am J Neuroradiol 19: 17271732.
32. Mormino EC, Kluth JT, Madison CM, Rabinovici GD, Baker SL, et al. (2009)
Episodic memory loss is related to hippocampal-mediated beta-amyloid
deposition in elderly subjects. Brain 132: 13101323.
33. Hansson O, Zetterberg H, Buchhave P, Londos E, Blennow K, et al. (2006)
Association between CSF biomarkers and incipient Alzheimers disease in
patients with mild cognitive impairment: a follow-up study. Lancet Neurol 5:
228234.
34. Ministe`re de la Sante (2009) Tarifs conventionnels des consultations de medecins
specialistes. In: Sante Mdl, ed. Paris: Assurance Maladie, Ministe`re de la Sante.
35. Ministe`re de la Sante (2009) Classification Commune des Actes Medicaux. In:
Sante Mdl, ed. Paris: Assurance Maladie, Ministe`re de la Sante.
36. Ministe`re de la Sante (2009) Nomenclature des actes de biologie medicale. In:
Sante Mdl, ed. Paris: Assurance Maladie, Ministe`re de la Sante.
37. Dictionnaire VIDAL. Paris: Editions du Vidal.
38. Feldman H, Gauthier S, Hecker J, Vellas B, Hux M, et al. (2004) Economic
evaluation of donepezil in moderate to severe Alzheimer disease. Neurology 63:
644650.
39. Rigaud AS, Fagnani F, Bayle C, Latour F, Traykov L, et al. (2003) Patients with
Alzheimers disease living at home in France: costs and consequences of the
disease. J Geriatr Psychiatry Neurol 16: 140145.
40. Elliott AF, Burgio LD, Decoster J (2010) Enhancing caregiver health: findings
from the resources for enhancing Alzheimers caregiver health II intervention.
J Am Geriatr Soc 58: 3037.
41. Reimat A (2002) Production associative et be nevolat informel : quelle
signification economique pour les activites de production des retraites ?
Innovations 15: 7398.
42. Marbot C (2008) Travailler pour des particuliers : souvent une activite
dappoint. Insee References, Les salaires en France.
43. France Benevolat (2010) La situation du benevolat en France en 2010. IFOP France Benevolat.
44. Fryback DG, Dasbach EJ, Klein R, Klein BE, Dorn N, et al. (1993) The Beaver
Dam Health Outcomes Study: initial catalog of health-state quality factors. Med
Decis Making 13: 89102.
45. Gagnon M, Rive B, Hux M, Guilhaume C (2007) Cost-effectiveness of
memantine compared with standard care in moderate-to-severe Alzheimer
disease in Canada. Can J Psychiatry 52: 519526.
46. Lejeune C, Bismuth MJ, Conroy T, Zanni C, Bey P, et al. (2005) Use of a
decision analysis model to assess the cost-effectiveness of 18F-FDG PET in the
Alzheimers Association (2009) 2009 Alzheimers disease facts and figures.

Alzheimers Dement 5: 234270.
Ballard C, Gauthier S, Corbett A, Brayne C, Aarsland D, et al. (2011)
Alzheimers disease. Lancet 377: 10191031.
Ferri CP, Prince M, Brayne C, Brodaty H, Fratiglioni L, et al. (2005) Global
prevalence of dementia: a Delphi consensus study. Lancet 366: 21122117.
Helmer C, Pasquier F, Dartigues JF (2006) [Epidemiology of Alzheimer disease
and related disorders]. Med Sci (Paris) 22: 288296.
Hampel H, Broich K, Hoessler Y, Pantel J (2009) Biological markers for early
detection and pharmacological treatment of Alzheimers disease. Dialogues Clin
Neurosci 11: 141157.
Forlenza OV, Diniz BS, Gattaz WF (2010) Diagnosis and biomarkers of
predementia in Alzheimers disease. BMC Med 8: 89.
McKhann GM (2011) Changing concepts of Alzheimer disease. JAMA 305:
24582459.
Weiner MW, Aisen PS, Jack CR, Jr., Jagust WJ, Trojanowski JQ, et al. (2010)
The Alzheimers disease neuroimaging initiative: progress report and future
plans. Alzheimers Dement 6: 202211 e207.
Kulasingam SL, Samsa GP, Zarin DA, Rutschmann OT, Patwardhan MB, et
al. (2003) When should functional neuroimaging techniques be used in the
diagnosis and management of Alzheimers dementia? A decision analysis. Value
Health 6: 542550.
McMahon PM, Araki SS, Neumann PJ, Harris GJ, Gazelle GS (2000) Costeffectiveness of functional imaging tests in the diagnosis of Alzheimer disease.
Radiology 217: 5868.
McMahon PM, Araki SS, Sandberg EA, Neumann PJ, Gazelle GS (2003) Costeffectiveness of PET in the diagnosis of Alzheimer disease. Radiology 228:
515522.
Furiak NM, Klein RW, Kahle-Wrobleski K, Siemers ER, Sarpong E, et al.
(2010) Modeling screening, prevention, and delaying of Alzheimers disease: an
early-stage decision analytic model. BMC Med Inform Decis Mak 10: 24.
Moatti JP, Le Corroller Soriano AG, Protiere C (2003) The Plan Cancer in
France: an economists point of view. Bull Cancer 90: 10101015.
Tan-Torres Edjerer T, Baltussen R, Adam T, Hutubessy R, Acharya A, et al.
(2003) WHO guide to cost-effectiveness analysis. Geneva: World Health
Organization.
TreeAge Software I (2009) TreeAge Pro Healthcare. Williamstown, USA.
Genin E, Hannequin D, Wallon D, Sleegers K, Hiltunen M, et al. (2011) APOE
and Alzheimer disease: a major gene with semi-dominant inheritance. Mol
Psychiatry 16: 903907.
Neumann PJ, Hermann RC, Kuntz KM, Araki SS, Duff SB, et al. (1999) Costeffectiveness of donepezil in the treatment of mild or moderate Alzheimers
disease. Neurology 52: 11381145.
Fagnani F, Lafuma A, Pechevis M, Rigaud AS, Traykov L, et al. (2004)
Donepezil for the treatment of mild to moderate Alzheimers disease in France:
the economic implications. Dement Geriatr Cogn Disord 17: 513.
Himes CL, Wagner GG, Wolf DA, Aykan H, Dougherty DD (2000) Nursing
home entry in Germany and the United States. J Cross Cult Gerontol 15:
99118.
INED (2008) Age- and sex-specific mortality rates, France, 2008. Paris, .
Winblad B, Engedal K, Soininen H, Verhey F, Waldemar G, et al. (2001) A 1year, randomized, placebo-controlled study of donepezil in patients with mild to
moderate AD. Neurology 57: 489495.
Rogers SL, Farlow MR, Doody RS, Mohs R, Friedhoff LT (1998) A 24-week,
double-blind, placebo-controlled trial of donepezil in patients with Alzheimers
disease. Donepezil Study Group. Neurology 50: 136145.
McShane R, Areosa Sastre A, Minakaran N (2006) Memantine for dementia.
Cochrane Database Syst Rev. CD003154 p.
Reisberg B, Doody R, Stoffler A, Schmitt F, Ferris S, et al. (2003) Memantine in
moderate-to-severe Alzheimers disease. N Engl J Med 348: 13331341.
13

47.
48.
49.
50.
management of metachronous liver metastases of colorectal cancer. J Nucl Med

46: 20202028.
Faissol DM, Griffin PM, Swann JL (2009) Bias in Markov models of disease.
Math Biosci 220: 143156.
Cohen JT, Neumann PJ (2008) Decision analytic models for Alzheimers disease:
state of the art and future directions. Alzheimers Dement 4: 212222.
Villemagne VL, Pike KE, Darby D, Maruff P, Savage G, et al. (2008) Abeta
deposits in older non-demented individuals with cognitive decline are indicative
of preclinical Alzheimers disease. Neuropsychologia 46: 16881697.
Forsberg A, Engler H, Almkvist O, Blomquist G, Hagman G, et al. (2008) PET
imaging of amyloid deposition in patients with mild cognitive impairment.
Neurobiol Aging 29: 14561465.
51. Wolk DA, Price JC, Saxton JA, Snitz BE, James JA, et al. (2009) Amyloid
imaging in mild cognitive impairment subtypes. Ann Neurol 65: 557568.
52. Reese JP, Hessmann P, Seeberg G, Henkel D, Hirzmann P, et al. (2011) Cost
and Care of Patients with Alzheimers Disease: Clinical Predictors in German
Health Care Settings. J Alzheimers Dis.
53. Mauskopf J, Mucha L (2011) A review of the methods used to estimate the cost
of Alzheimers disease in the United States. Am J Alzheimers Dis Other Demen
26: 298309.
54. Schachter F, Faure-Delanef L, Guenot F, Rouger H, Froguel P, et al. (1994)
Genetic associations with human longevity at the APOE and ACE loci. Nat
Genet 6: 2932.
55. Caselli RJ, Walker D, Sue L, Sabbagh M, Beach T (2010) Amyloid load in
nondemented brains correlates with APOE e4. Neurosci Lett 473: 168171.
29
14
Multi-Method Analysis of MRI Images in Early

Diagnostics of Alzheimers Disease
Robin Wolz1., Valtteri Julkunen2., Juha Koikkalainen3, Eini Niskanen2,4, Dong Ping Zhang1, Daniel
Rueckert1, Hilkka Soininen2,5, Jyrki Lotjonen3*, the Alzheimers Disease Neuroimaging Initiative"
1 Biomedical Image Analysis Group, Department of Computing, Imperial College London, London, United Kingdom, 2 Department of Neurology, Kuopio University
Hospital, Kuopio, Finland, 3 Knowledge Intensive Services, VTT Technical Research Centre of Finland, Tampere, Finland, 4 Department of Applied Physics, University of
Eastern Finland, Kuopio, Finland, 5 Institute of Clinical Medicine, Neurology, University of Eastern Finland, Kuopio, Finland
Abstract
The role of structural brain magnetic resonance imaging (MRI) is becoming more and more emphasized in the early
diagnostics of Alzheimers disease (AD). This study aimed to assess the improvement in classification accuracy that can be
achieved by combining features from different structural MRI analysis techniques. Automatically estimated MR features
used are hippocampal volume, tensor-based morphometry, cortical thickness and a novel technique based on manifold
learning. Baseline MRIs acquired from all 834 subjects (231 healthy controls (HC), 238 stable mild cognitive impairment (SMCI), 167 MCI to AD progressors (P-MCI), 198 AD) from the Alzheimers Disease Neuroimaging Initiative (ADNI) database
were used for evaluation. We compared the classification accuracy achieved with linear discriminant analysis (LDA) and
support vector machines (SVM). The best results achieved with individual features are 90% sensitivity and 84% specificity
(HC/AD classification), 64%/66% (S-MCI/P-MCI) and 82%/76% (HC/P-MCI) with the LDA classifier. The combination of all
features improved these results to 93% sensitivity and 85% specificity (HC/AD), 67%/69% (S-MCI/P-MCI) and 86%/82% (HC/
P-MCI). Compared with previously published results in the ADNI database using individual MR-based features, the presented
results show that a comprehensive analysis of MRI images combining multiple features improves classification accuracy and
predictive power in detecting early AD. The most stable and reliable classification was achieved when combining all
available features.
Citation: Wolz R, Julkunen V, Koikkalainen J, Niskanen E, Zhang DP, et al. (2011) Multi-Method Analysis of MRI Images in Early Diagnostics of Alzheimers
Disease. PLoS ONE 6(10): e25446. doi:10.1371/journal.pone.0025446
Editor: Celia Oreja-Guevara, University Hospital La Paz, Spain
Received July 12, 2011; Accepted September 5, 2011; Published October 13, 2011
Copyright: 2011 Wolz et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
Funding: This project is partially funded under the 7th Framework Programme by the European Commission (http://cordis.europa.eu/ist/). Data collection and
sharing for this project was funded by the Alzheimers Disease Neuroimaging Initiative (ADNI; Principal Investigator: Michael Weiner; NIH grant U01 AG024904).
ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering (NIBIB), and through generous contributions
from the following: Pzer Inc., Wyeth Research, Bristol-Myers Squibb, Eli Lilly and Company, GlaxoSmithKline, Merck & Co. Inc., AstraZeneca AB, Novartis
Pharmaceuticals Corporation, Alzheimers Association, Eisai Global Clinical Development, Elan Corporation, plc, Forest Laboratories, and the Institute for the Study
of Aging, with participation from the U.S. Food and Drug Administration. Industry partnerships are coordinated through the Foundation for the National Institutes
of Health. The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimers Disease
Cooperative Study at the University of California San Diego. ADNI data are disseminated by the Laboratory of Neuro Imaging at the University of California Los
Angeles. Data used in the preparation of this article were obtained from the ADNI database (www.loni.ucla.edu/ADNI). As such, the investigators within the ADNI
contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. ADNI investigators
include (complete listing available at http://adni.loni.ucla.edu/wp-content/uploads/how_to_apply/ADNI_Authorship_List.pdf).
. These authors contributed equally to this work.
" For information on the Alzheimers Disease Neuroimaging Initiative please see the Acknowledgments section.
Introduction
The search for reliable biomarkers of AD-type pathology and

predictors of disease progression among MCI subjects is ongoing.
AD is characterized by neurofibrillary tangles and amyloid plaques
in the brain [5]. Degenerative changes in the human neurotransmitter system lead to atrophy in selected brain regions [6]. The
most promising candidate biomarkers are the ones derived from
structural and functional neuroimaging as well as those measured
in cerebrospinal fluid (CSF) and plasma [7]. Amyloid-based
measures like the CSF-peptide Ab42 and the uptake of the PiB
tracer on positron emission imaging (PET) show the earliest ADtype changes [7]. However, there is evidence that the number of
amyloid plaques reach their saturation levels already by the time
patients have clinically apparent symptoms of cognitive impairment [8,9], whereas atrophy, neuronal loss, synaptic loss, and the
number of tangles increase with severity of illness [10]. These
Alzheimers disease (AD) is the most common cause of dementia

globally and one of the major healthcare issues of the future. It has
been estimated that during the next four decades the prevalence of
AD will quadruple from 27 to 106 million by which time 1 in 85
persons worldwide will be living with the disease [1]. Even a
modest delay of one year in disease onset and progression could
reduce the number of cases by 9 million [1]. Interventions are
postulated to be most effective when directed at patients at the
earliest stages of the disease, which underlines the importance of
early diagnosis of AD [2]. Mild cognitive impairment (MCI) is a
heterogeneous syndrome that increases the risk of developing AD
markedly [3]. However, not all MCI subjects convert to AD and
some may even return to normal cognition [4].
30
October 2011 | Volume 6 | Issue 10 | e25446

Early Diagnostics of AD Using MRI-Based Biomarkers
Materials and Methods
findings suggest that, although amyloid-based biomarkers may be

used as longitudinal markers of AD type pathology, they seem to
offer only limited insight into which MCI subjects will most likely
convert to AD in the near future. In a recently published dynamic
model of biomarker behavior in the AD spectrum, biomarkers
based on structural magnetic resonance imaging (MRI) have been
shown to be correlated with a progression from MCI to AD [11].
Such biomarkers could therefore improve the accuracy of early
AD diagnostics and reduce especially the amount of false positive
diagnoses. Besides providing chance for a more focused and earlier
intervention, structural MRI biomarkers of AD could also aid the
development of new disease-modifying drugs by acting as surrogate
markers of disease progression, reduce the number of subjects
needed to detect significant drug effect and provide quantitative
measures of treatment benefits [12].
It has been shown that the early diagnostics of AD can be
improved by using multiple different biomarkers simultaneously.
Usually these studies have combined MRI-based markers with
biomarkers based on positron emission tomography (PET) [13,14],
cerebrospinal fluid (CSF) [15,16] or both [1719]. Achieved
results vary from no additional benefit [15,17] to significant
improvement [13,14,16,20]. However, availability of all three
biomarkers (CSF, PET, MRI) is not very common in clinical
practice since obtaining all measures is laborious for the patient
and clinician, induces delays and increases the costs of the
diagnosis significantly. Furthermore, measurements obtained from
CSF and PET are considered invasive. Recent studies focusing on
only structural MRI have reached correct classification accuracys
(CCR) of 7694% in identifying healthy controls (HC) from
patients with AD and 6482% in predicting which MCI subjects
will convert to AD in the imminent future [2127]. The high
variation in these results can be attributed to differences in study
populations as well as evaluation designs. With the Alzheimers
Disease Neuroimaging Study (ADNI) [28], a large multi-center
study on MR imaging in AD has been established that is available
to the wider research community. Based on a large sub-group of
ADNI subjects, Cuingnet et al. [29] presented a comparison of ten
MRI-based feature extraction methods and their ability to
discriminate between clinically relevant subject groups. The ten
methods evaluated comprise five voxel-based methods, three
methods based on cortical thickness and two methods based on the
hippocampus. Best sensitivity/specificity values reported are 81%/
95% for AD vs HC, 70%/61% for S-MCI vs P-MCI and 73%/
85% for HC vs P-MCI.
In this paper we use the ADNI database to evaluate the ability
of the combination of different MR-based features to increase
classification accuracy. We evaluate the power of hippocampal
volume (HV), cortical thickness (CTH), tensor-based morphometry (TBM) and features extracted from a recently proposed
manifold-based learning (MBL) framework to discriminate healthy
controls from subjects with AD and to predict conversion from
MCI to AD. For evaluation we used all 834 ADNI baseline images
that were available from the ADNI webpage. Compared to
previous work this paper aims at establishing the improvement in
accuracy and stability that can be achieved by combining more
than one MR-based feature. To the best of our knowledge it is the
first comprehensive study that analyzes MRI-derived features for
the full ADNI dataset. For direct comparison with the work by
Cuingnet et al. [29] we also evaluated all results on the subset used
in their work.
To test the influence of the classification method used, we utilized
both support vector machines (SVMs) and a linear discriminant
analys (LDA) to evaluate classification accuracy (CCR), sensitivity
(SEN) and specificity (SPE) in each experiment.
Subjects
In the ADNI study, brain MR images were acquired at regular
intervals after an initial baseline scan from approximately 200
cognitively normal older subjects (HC), 400 subjects with mild
cognitive impairment (MCI), and 200 subjects with early AD.
Detailled inclusion/exclusion criteria used for the different subject
groups in ADNI are defined in [30]. The AD group has scores
between 2026 (inclusive) on the Mini-Mental State Examination
(MMSE) [31], and a Clinical Dementia Rating (CDR) [32] of 0.5
or 1.0. Furthermore, these subjects fulfil the NINCDS/ADRDA
criteria for probable AD [33]. MCI subjects included have MMSE
scores between 2430 (inclusive), a memory complaint, have
objective memory loss measured by education adjusted scores on
Wechsler Memory Scale Logical Memory II, a CDR of 0.5,
absence of significant levels of impairment in other cognitive
domains, essentially preserved activities of daily living, and an
absence of dementia [30]. Healthy subjects have MMSE scores
between 2430 (inclusive), a CDR of 0, are non-depressed, non
MCI, and nondemented. A more detailed description of the ADNI
study is given in Appendix S1.
All 834 ADNI subjects (231 HC, 238 S-MCI, 167 P-MCI, 198
AD) for which a 1.5T T1-weighted MRI scan at baseline was
available were included in this study. 167 subjects in the MCI
group converted to AD as of July 2011. We therefore
independently analysed progressive MCI (P-MCI) subjects and
subjects with a stable diagnosis of MCI (S-MCI).
Table 1 shows the demographics for the 834 study subjects.
Statistically significant differences in the demographics and clinical
variables between the study groups were assessed using Students
unpaired t-test. In this work, the difference was considered
statistically significant if pv0.05 if not stated otherwise. There
were more men than women in all other groups besides the AD
group. MMSE scores were significantly different in the pairwise
comparisons between all study groups. CDR scores of the HC and
AD groups are significantly different to the ones of the two MCI
groups. Healthy subjects had a significantly lower Geriatric
Depression Scale (GDS) compared to all other groups. Compared
to all other groups, AD subjects had significantly shorter education.
MRI Acquisition
Standard 1.5T screening/baseline T1-weighted images obtained using volumetric 3D MPRAGE protocol with resolutions
Table 1. Subjects.
Group
HC
S-MCI
P-MCI
AD
231
238
167
198
Men
52%
66%
62%
52%
Age
76.02 (5.0)
74.85 (7.8)
74.6 (7.0)
75.68 (7.7)
MMSE
29.1* (1.0)
27.3* (1.8)
26.6* (1.7)
23.3* (2.0)
CDR
0 (0)
0.49 (0.05)
0.50 (0)
0.75 (0.25)
GDS
0.83* (1.14)
1.60 (1.42)
1.53 (1.30)
1.67 (1.42)
Education
16.0 (2.8)
15.6 (3.1)
15.7 (2.9)
14.7* (3.1)
APOE4 status
(e3e4/e4e4)
23%/2%
31%/8%
50%/16%
42%/18%
Months to
conversion
18.2 (10.1)
*means statistically significant different from all other groups.

31
Instead of using just one template image, we used 30 randomly

selected images (10 controls, 10 MCIs, and 10 ADs) from the
ADNI database as template images. The template images were
used also in the classification analysis to maximize the number of
subjects. Each template image was registered to a study image, and
Jacobian maps were computed for each template image. To
combine the results of multiple templates, all template images were
registered to the mean anatomical template generated from the 30
images, and all the results were normalized to this reference space
[47]. The combination of the results was performed by averaging
the ROI-wise feature values of all the templates as described in
detail below.
Manifold-based learning. In this machine learning
approach, non-linear dimensionality reduction with Laplacian
eigenmaps [48] is used to learn features to discriminate between
different subject groups. Laplacian eigenmaps estimates the lowdimensional representation of a set of input images based on a
similarity graph that is defined with pairwise image similarities [48].
The hypothesis is that such a low-dimensional representation
captures the variability in the dataset in a more compact way than
pairwise image similarities directly. We estimate pairwise image
similarities from the intensity appearance in a region around
hippocampus and amygdala since both structures are known to be
affected by AD in an early stage. All images are aligned in a
template space using a coarse non-rigid registration (10 mm Bspline control-point spacing, [49]). Such a coarse non-rigid
alignment ensures that corresponding brain structures are aligned
but still allows to measure subject-specific differences. After
performing dimensionality reduction, the first 20 dimensions of
the resulting manifold are used as features to perform classification
with the different methods used. More details on the theory and
application of this manifold learning approach can be found in
[20,50]. Figure 1 exemplarily shows a 2D embedding of a set of
ADNI images acquired from healthy controls and subjects with AD.
It can be seen that even two embedding dimensions give a relatively
good separation between both groups. In our experiments we used a
higher dimensional space allowing better discrimination.
ranging from 0.9 mm|0.9 mm|1.20 mm to 1.3 mm|1.3 mm

|1.20 mm were included from the ADNI database. For detailed
information of the MRI protocols and preprocessing steps see [34].
Feature extraction
32
All fully automated feature extraction methods described below

were applied to images that were preprocessed by the ADNI
pipeline.
Hippocampal volume. Baseline hippocampal volumes
were measured using an approach based on fast and robust
multi-atlas segmentation [35,36]. In this approach, multi-atlas
label propagation is applied in combination with atlas selection to
obtain the hippocampus segmentation. A set of hippocampus
atlases is selected from a pool of atlas images according to image
similarity with the query image. After registering all atlases to the
query image, a spatial prior is generated from the multiple label
maps. This spatial prior is then used to obtain a final segmentation
based on an expectation maximization (EM) segmentation
algorithm [37].
Cortical thickness. CTH is measured in the baseline
T1-weighted structural MR images by using an automated
computational surface-based method developed at the McConnell
Brain Imaging Centre, Montreal Neurological Institute, McGill
University, Montreal, Canada (http://www2.bic.mni.mcgill.ca/)
[38]. Individual MRI volumes were registered to standard space
using the ICBM152 template [39]. Intensity non-uniformities were
corrected [40] before the final brain mask was calculated [41].
Tissues were segmented into white matter (WM), grey matter (GM)
and cerebrospinal fluid (CSF) using the INSECT-algorithm [42]
and the magnitude of PVE was estimated by using the trimmed
minimum covariance determinant (TMCD) method [43]. The
brains were divided automatically into two separate hemispheres
and the inner and outer surfaces of the cortex were extracted
according to intersections between WM and GM (white matter
surface, WMS) as well as GM and CSF (grey matter surface, GMS)
using the Constrained Laplacian-Based Automated Segmentation
with Proximities (CLASP) algorithm [44]. The inner surface was
first formed by deforming an ellipsoid polygon mesh to the shape of
the WMS. GMS was obtained by further expanding the inner
surface. Each polygon mesh surface consisted of 81,920 polygons
and 40,962 nodes per hemisphere. The thickness of the cortex was
defined at each linked node as the distance between the two
concentrically linked polygon meshes on the WMS and the GMS.
This t-link metric has been proven to be the simplest yet most
precise way to determine cortical thickness [38]. Although MR
images were transformed to standard space to allow for group
analysis, thickness calculations were performed in each subjects
native space. Finally, cortical thickness maps were smoothed with a
20 mm FWHM diffusion smoothing kernel to improve the signalto-noise ratio and statistical power [45]. The described toolbox did
not achieve satisfactory results on some study subjects because of i)
failure in tissue segmentation and brain masking (48 subjects) and ii)
failure in partial volume effect estimation (59 subjects). As a result
the pipeline crashed and CTH measures were not obtained for 76
subjects (24 control, 35 MCI, 17 AD). Also the cortical model of 31
subjects (10 control, 13 MCI, 8 AD) was completely deformed and
thus unusable. For these 107 subjects the CTH features were
considered as missing values. CTH features used in the classification
experiments are introduced below.
Tensor-based morphometry. The TBM analysis was
performed using a multi-template approach [46,47]. In TBM, a
template image is non-rigidly registered to a study image, and,
typically, the determinant of the Jacobian matrix (the Jacobian) of
the deformation is used to measure the voxel-level morphometry.
ROI-wise features for CTH and TBM

Both CTH and TBM analyses produce local (point-wise)
information, either on cortical thickness or the volume. Thus,
the number of original features is enormous, and to make the
classification more efficient and robust, the number of features has
to be reduced. We evaluated both features in a statistical region of
interest (ROI) defined as detailed in Appendix S2. Figures 2 and 3
show t-values for statistically significant differences between study
groups for TBM and CTH respectively. A detailed description of
the definition of these statistical ROIs is given in Appendix S2.
Study design
Table 2 presents an overview on the features calculated for all
834 available ADNI baseline images. All feature values were
corrected for age and gender using a linear regression model
where control subjects were used as the training set, i.e., the
normal, not disease-related, age and gender related differences in
the classification features were removed. Feature selection was
then carried out on the corrected feature sets using stepwise
regression [51].
We used two subsets to perform classification:
I.
II.
All 834 available baseline images described in the subjects

section
509 baseline images used by Cuingnet et al. [29] and detailed
in their publication.

Figure 1. 2D manifold embedding of a set of images acquired from healthy controls (red) and subjects with AD (blue).
Consequently, in overall, the classification evaluation was

performed using 300 (3|100) repetitions, and the results
presented in this paper are the average values of all these
classifications.
Dataset II. Statistical ROIs for CTH and TBM feature
extraction were calculated from the 325 baseline images that are
not part of dataset II. In order to allow direct comparison of
classification accuracy with the work by Cuingnet et al. [29],
separate training and testing sets for the different comparisons
were defined using the exact sub-groups reported in their
manuscript. Around 50% of all subjects are used to train the
different types of classifiers and the reported results are based on
classifying the remaining subjects.
The following sections describe the definition of the statistical

ROIs and evaluation strategy used for the two datasets
respectively.
Dataset I. In order to perform the study using crossvalidation in the full dataset, it was divided into three equally
sized parts. One part was used to perform the statistical tests for
the CTH and TBM features, and the remaining two parts were
used to evaluate the classification accuracy. This was repeated
three times so that each part was once used to perform the
statistical tests. Afterwards, the results of the three repetitions were
averaged. The classification accuracy was evaluated using leave-Nout cross validation on those subjects not included in the statistical
tests. Five percent of the evaluation subjects were regarded as the
test set, and the remaining 95% of the subjects were used to train a
classifier which was then applied to the test set. This was repeated
table-1-caption100 times, each time selecting randomly the test set
subjects. Finally, the results of the 100 repetitions were averaged.
Classification methods
We used two different widely used methods to perform
classification based on individual features and their combination:
Figure 2. Results for voxelwise t-tests for statistically significant group differences with features extracted from TBM.
33
Figure 3. Results for t-tests for statistically significant group differences based on cortical thickness measurements.
Linear discriminant analysis (LDA). Linear discriminant

analysis (LDA) is a widely used technique to find a linear
combination of features to best separate several classes [52]. In this
work we used LDA as implemented in the classify function in
Matlab with a multivariate normal density model with uninformative priors (p = 0.5).
Support vector machines (SVM). Support vector machines
use training data to find a separating hyperplane in the ndimensional training space that best separates two subject groups
[53]. Test subjects are then classified according to their position
relative to the defined hyperplane in the n-dimensional feature
space. We used the libSVM library to perform the analysis. The
radial basis function kernel was selected based on the guidelines
provided by the libSVM library (Software available 2.3.2011 at
http://www.csie.ntu.edu.tw/cjlin/libsvm).
significant improvements achieved when combining all features are

marked with { (pv0.0001). To test for significance, unpaired t-tests
were carried out between distribution estimates for the corresponding classification rates based on the multiple runs. All estimated
distributions passed a normality test using a Kolmogorov-Smirnov
test at a~0:05.
For direct comparison with work presented by Cuingnet et al.
[29], we performed classification based on the training- and testing
sets defined in their manuscript as described above. S-MCI and PMCI groups are defined in the same way as in the original
publication. Sensitivity and specificity values for the classification
in all three clinical pairings are reported in Table 6. Following the
clear advantage for LDA in the performance on the full dataset,
we only report results with this classifier for dataset II.
Discussion
Results
In this study we assessed the automatic diagnostic capabilities of

4 structural MRI features (MBL, HC, CTH, TBM) separately and
combined in 834 baseline images acquired in the ADNI study.
When applied separately, TBM provided the overall best results,
closely followed by MBL. Combining all features improved
the results in all study experiments. Our results show how a
combination of different MRI-based features can improve results
based on only one measurement, resulting in a more powerful and
stable classifier. The most significant improvement of the combination
We used both classification methods to measure classification

accuracy based on individual features as well as the combination of
all features. The results for the comparisons HC vs AD, HC vs PMCI and S-MCI vs P-MCI in the full ADNI database are presented
in Tables 3, 4 and 5 respectively. Presented are classification
accuracy (CCR), sensitivity (SEN) and specificity (SPE). Furthermore, the 95% confidence interval for the classification accuracy is
estimated based on the multiple classification runs. Statistically
Table 2. Features used in the study.
Method
No of features
Description
Hippocampal volume (HV)
total volume of left and right hippopcampus

average cortical thickness within a ROI defined based on group-level statistical analysis
Cortical thickness
9 (HC vs AD)
(CTH)
7 (HC vs P-MCI)
Tensor-based morphometry (TBM)
84
average Jacobian of atrophic voxels within a ROI, weighted based on voxel-wise p-values
Manifold-based learning (MBL)
20
coordinates of a subject in a low-dimensional manifold space learned from pairwise image

similarities
8 (S-MCI vs P-MCI)
34

Table 3. Classification results for HC vs AD.
Feature
LDA
Table 5. Classification results for S-MCI vs P-MCI.
SVM
Feature
CCR [95% CI]
SEN
SPE
CCR [95% CI]
SEN
SPE
MBL
85{ [64 100]
87
83
85 [64 100]
87
83
HV
81{ [57 100]
81
79
81{ [57 100]
84
77
CTH
81{ [64 100]
89
71
82{ [57 100]
90
TBM
87{ [71 100]
90
84
87 [71 100]
89
All
89 [71 100]
93
85
86 [71 100]
94
SEN
SPE
CCR [95% CI]
MBL
65{ [36 86]
64
66
65{ [43 86]
77
48
HV
65{ [36 86]
63
67
62 [36 86]
83
33
73
CTH
56{ [29 86]
63
45
59 [36 79]
96
03
84
TBM
64{ [36 86]
65
62
64{ [36 86]
77
44
78
All
68 [43 93]
67
69
60 [36 86]
92
14
SEN
SPE
{means statistically significant different from the combined results with

pv0.0001.
over the best individual feature can be observed for HC vs P-MCI

with 5% units followed by 3 and 2% units for S-MCI vs P-MCI and
HC vs AD, respectively. These improvements lead to 20, 12 and 9
subjects more being correctly classified respectively when using the
combined feature set as compared to the best single feature for every
comparison. Comparing two classification approaches based on LDA
and SVMs resulted in a clear advantage of the former.
Several studies reported classification results using single MRI
methods for the HC/AD classification (Table 7). Liu et al. [24]
reported SEN/SPE of 92/90 in the classification of HC/AD
subjects using regional cortical volumes in the AddNeuroMed
dataset. McEvoy et al. [26] report a CCR of 89 on images from
the ADNI database using features from cortical thickness and
structural volumes. Vemuri et al. [54] present a SEN/SPE of 86/
86 on 380 subjects using the STAND score. In our study the
results obtained with single methods are lower (7190) but almost
identical when the methods were combined. It should be noted,
however, that Liu and colleagues did not use cross-validation or
separate training/testing sets when producing the results which
could lead to overestimation of the results in a dataset outside the
study cohort. Gerardin et al. [23] acquired a high SEN/SPE of
96/92 by using hippocampal shape analysis, but the number of
subjects (25 HC, 23 AD) was quite low in order to produce results
with good generalizability. Westman et al. [55] reported a CCR of
82 for HC vs AD classification and 73 for HC vs P-MCI
classification by using various regional brain volumes. Our results
are substantially more accurate, the group sizes are larger and
clinical follow-up time is one year longer. Chupin et al. [21]
reported SEN/SPE of 75/77 (hippocampal volume) and Querbes
et al. [27] a CCR of 85 (cortical thickness), both lower than the
results acquired with the combination of features or TBM features

independently in our study.
Varying results concerning AD prediction (S-MCI/P-MCI
classification using baseline measurements) have been published
(Table 7): Querbes et al. [27] reported a CCR of 73, Liu et al. [25]
a SEN/SPE of 76/68, Chupin et al. [21] reported a SEN/SPE of
60/65 and Davatzikos et al. [15] SEN/SPE of 95/38. Our results
with separate and combined baseline features lie in the range of
these results (SEN/SPE 63/67, 64/66 and 67/69 when using HV,
MBL and the combined features, respectively).
There can be several explanations for the variation in the
reported results. A majority of the studies in this field have used
different statistical methods and MRI feature extraction strategies
on different datasets, which makes a comparison of the results
complicated. Also the variation in the size of the study samples and
the use (or ignoring) of cross-validation or separate training/testing
sets are important factors, which both have crucial impact on the
reliability and generalizability of the results. In Lotjonen et al.
[36], we demonstrated that choosing from a population of 350
cases several times 2/3 for the training set and 1/3 for the test set
and using hippocampus volume as a classification feature can lead
to any classification accuracy between 53% and 77%. This
observation is also confirmed by the high confidence intervals for
the classification accuracies reported in Tables 3, 4 and 5. This
shows that a fair comparison of methods based on the classification
accuracy is difficult if not exactly the same data and classification
approaches are used. Furthermore, since the ADNI study is still
ongoing, several subjects labeled as S-MCI will progress in the
future to the P-MCI group.
A recent study with a subset of ADNI subjects assessed the
classification performance of several structural MRI methods in
experiments comparable to our investigation [29]. Reported
Table 4. Classification results for HC vs P-MCI.
LDA
SVM
CCR [95% CI]

pv0.0001. CCR = Correct classification rate, SEN = Sensitivity, SPE = Specificity.
Feature
LDA
Table 6. Classification results based on a subset of ADNI that

was previously used for classification by Cuingnet et al. [29].
SVM
CCR [95% CI]
SEN
SPE
CCR [95% CI]
SEN
SPE
MBL
78{ [54 100]
81
75
77{ [54 92]
84
69
HV
76{ [54 92]
77
76
78 [54 92]
83
71
CTH
77{ [54 100]
85
65
77 [54 100]
89
62
TBM
79{ [62 100]
82
76
80{ [62 100]
85
74
All
84 [62 100]
86
82
82 [62 100]
93
67
Feature
SEN

pv0.0001.
HC vs AD
SPE
HC vs P-MCI
S-MCI vs P-MCI
SEN
SEN
SPE
SPE
MBL
90
74
84
92
55
76
HV
80
69
75
76
63
70
CTH
85
75
86
59
72
35
TBM
93
76
90
84
63
59
All
94
76
94
89
69
54
35
Table 7. Classification results of healthy control (HC), mild cognitive impairment (MCI) and Alzheimers disease subjects reported
in the recent literature.
Study
Features
HC vs AD
HC vs P-MCI
S-MCI vs P-MCI
CCR
SEN
SPE
CCR
SEN
SPE
CCR
SEN
SPE
Liu et al. [24]
333
Cortical volumes
91
92
90
Gerardin et al. [23]*
70
Hippocampus shape
94
96
92
Chupin et al. [21]*
605
Hippocampus volume
76
75
77
64
60
65
Querbes et al. [27]*
382
Cortical thickness
85
73
75
68
Liu et al. [25]
312
Amygdala/caudate volumes
69
76
68
Davatzikos et al. [15]*
356
SPARE-AD index
56
95
38
Cuingnet et al. [29]*
509
Various
81
95
73
85
62
69
Hinrichs et al. [14]*
159
MRI & PET
81
60
92
14
Westman et al. [55]
351
Various volumes
82
73
McEvoy et al. [26]*
398
Cortical thickness/various volumes
89
83
93
Vemuri et al. [54]
380
STAND score
86
86
N = Number of study subjects,

* = ADNI dataset.
36
Considering solely the classification accuracies of the present

study and those reported in literature, it seems questionable if the
collection of several biomarkers is worth the effort and resource. A
combination of different features extracted from a single MRI
seems to provide results that are comparable or better than those
obtained with other or multiple biomarkers. In a clinical point of
view, this is interesting since it means that a single MRI scan
provides not only aid to differential diagnostics of cognitive
impairment, but also reliably describes a persons phase in the HC/
AD continuum. MRI is also widely available, non-invasive and
often useful in the differential diagnostics of memory problems
thus making it a compelling option as the first biomarker that
would be obtained from a patient with mild memory problems.
However, a comprehensive differential diagnostics between AD
and non-AD cognitive impairments will still require assessment of
various different biomarkers. Also, it should be noted that the
computational techniques used in this paper are not widely
available in the clinical environment and thus limit their usage in
the clinical work at present.
Strengths of the presented study are i) the use of multiple
features extracted from one imaging modality, ii) large groups, iii)
rigorous validation process of the results using cross-validation,
and iv) results comparable or better than the ones published so far.
Our study has also some limitations that should be mentioned.
The results are obtained from a single (although collected from
multiple sites) cohort and should be also validated in other cohorts.
A longer clinical follow-up time would be needed to see if the
classification results of S-MCI/P-MCI experiment changed when
more of the MCI subjects converted to AD. Furthermore, the
ADNI study does not provide postmortem pathological confirmation of the clinical status. With this limitation, individual subjects
might be wrongly categorized. Although a rigorous validation
process was used, optimally we need to establish standardized cutoffs that would be well generalizable to other cohorts outside
ADNI. That is, however, beyond the possibilities of this study and
will require vast standardization and validation procedures. Also,
the CTH pipeline had problems especially with severely atrophied
brains or MRI scans with poor image quality. A more robust
pipeline would be desirable in order to guarantee a more reliable
feature extraction.
SEN/SPE lie in the ranges 59/8181/95 (HC vs AD) and 70/73

73/85 (HC vs P-MCI). While most methods tested did not exceed
the accuracy of a random classifier for the discrimination between
S-MCI and P-MCI, the best results reported for this task were a
SEN/SPE of 62/69 when using hippocampal volume. To allow a
direct comparison of the results reported by Cuingnet et al. [29],
we evaluated our features on the exact same training- and testing
sets used in their paper. This direct comparison shows that our
results compare favourably to other, established methods in
neuroimaging. For HC vs AD classification, individual features in
our study give more sensitive but less specific results than most
methods in the previous publication. Combining all features gives
an overall better classification accuracy than the majority of
previously tested methods. Our results on the combined feature set
furthermore outperform the majority of methods tested by
Cuingnet et al. [29] when predicting MCI conversion as well as
all methods for the classification between HC and P-MCI. A
significant difference in classification accuracy can be observed
between the full ADNI dataset and this smaller subset used for
comparison with previous work. Reasons may include a strict
separation into trainin- and testing sets which may result in less
generalisability as well as the shorter follow-up period that was
considered to define progression to AD.
Some studies have also combined different biomarkers (CSF,
MRI, PET) with the idea of measuring different aspects of AD
pathology and thus improve the classification accuracy. Hinrichs
et al. [14] improved their HC/AD classification CCR by a few %
units to 81 by combining MRI and PET. Eckerstrom et al. [16]
studied the separation of a unified HC/S-MCI group from P-MCI
group with CSF proteins and manual hippocampal volumes. They
found CSF to be superior to MRI (SEN/SPE of 95/79 vs 86/66)
while the combination performed best (SEN/SPE 90/91).
However, it should be noted that the study sample in that
particular study was small (a total of 68 subjects) and neither crossvalidation or separate training/testing sets were used in order to
ensure good generalizability of the results. In Kohannim et al.
[17], the improvement from using multiple biomarkers was not
significant and Davatzikos et al. [15] reported marginal improvements which, however, may be related to the fact that results with
only one biomarker were not very good to begin with.

ADNI contributed to the design and implementation of ADNI and/or

provided data but did not participate in analysis or writing of this report.
ADNI investigators include (complete listing available at http://adni.loni.
ucla.edu/wp-content/uploads/how_to_apply/ADNI_Authorship_List.
pdf).
Appendix S1
The Alzheimers Disease Neuroimaging Initiative.
(DOCX)
Appendix S2 ROI-wise features for CTH and TBM.
(DOCX)
Acknowledgments
Conceived and designed the experiments: RW VJ JK DR HS JL.

Performed the experiments: RW VJ JK EN DPZ. Analyzed the data: RW
VJ JK. Contributed reagents/materials/analysis tools: RW JK DR JL.
Wrote the paper: RW VJ.
Data used in the preparation of this article were obtained from the ADNI
database (www.loni.ucla.edu/ADNI). As such, the investigators within the
References
24. Liu Y, Paajanen T, Zhang Y, Westman E, Wahlund LO, et al. (2009)
Combination analysis of neuropsychological tests and structural MRI measures
in differentiating AD, MCI and control groupsThe AddNeuroMed study.
Neurobiology of Aging In Press, Corrected Proof: -.
25. Liu Y, Paajanen T, Zhang Y, Westman E, Wahlund LO, et al. (2010) Analysis of
regional MRI volumes and thicknesses as predictors of conversion from mild
cognitive impairment to Alzheimers disease. Neurobiology of Aging 31:
13751385.
26. McEvoy L, Fennema-Notestine C, JC R, Hagler D, Holland D, et al. (2009)
Alzheimer disease: quantitative structural neuroimaging for detection and
prediction of clinical and structural changes in mild cognitive impairment.
Radiology 251: 1950205.
27. Querbes O, Aubry F, Pariente J, Lotterie JA, Dmonet JF, et al. (2009) Early
diagnosis of alzheimers disease using cortical thickness: impact of cognitive
reserve. Brain 132: 20362047.
28. Mueller SG, Weiner MW, Thal LJ, Petersen RC, Jack C, et al. (2005) The
Alzheimers Disease Neuroimaging Initiative. Neuroimaging Clinics of North
America 15: 869877.
29. Cuingnet R, Gerardin E, Tessieras J, Auzias G, Lehricy S, et al. (2010)
Automatic classification of patients with Alzheimers disease from structural
MRI: A comparison of ten methods using the ADNI database. NeuroImage In
Press, Corrected Proof: -.
30. Petersen RC, Aisen PS, Beckett LA, Donohue MC, Gamst AC, et al. (2010)
Alzheimers Disease Neuroimaging Initiative (ADNI): clinical characterization.
Neurology 74: 201209.
31. Folstein MF, Folstein SE, McHugh PR (1975) Mini-mental state: A practical
method for grading the cognitive state of patients for the clinician. Journal of
Psychiatric Research 12(3): 189198.
32. Morris J (1993) The Clinical Dementia Rating (CDR): current version and
scoring rules. Neurology 43: 24122414.
Clinical diagnosis of alzheimers disease: report of the nincds-adrda work group
under the auspices of department of health and human services task force on
alzheimers disease. Neurology 34: 939944.
34. Jack CR, Jr., Bernstein MA, Fox NC, Thompson P, Alexander G, et al. (2008)
The Alzheimers disease neuroimaging initiative (ADNI): MRI methods. Journal
of Magnetic Resonance Imaging 27: 685691.
35. Lotjonen JM, Wolz R, Koikkalainen JR, Thurfjell L, Waldemar G, et al. (2010)
Fast and robust multi-atlas segmentation of brain magnetic resonance images.
NeuroImage 49: 23522365.
36. Lotjonen J, Wolz R, Koikkalainen J, Julkunen V, Thurfjell L, et al. (2011) Fast
and robust extraction of hippocampus from mr images for diagnostics of
alzheimers disease. NeuroImage 56: 185196.
37. Van Leemput K, Maes F, Vandermeulen D, Suetens P (1999) Automated
model-based tissue classification of MR images of the brain. IEEE Transactions
on Medical Imaging 18: 897908.
38. Lerch J, Evans A (2005) Cortical thickness analysis examined through power
analysis and a population simulation. NeuroImage 24: 163173.
39. Mazziotta J, Toga A, Evans A, Fox P, Lancaster J, et al. (2001) A probabilistic
atlas and reference system for the human brain: International Consortium for
Brain Mapping (ICBM). Philos Trans R Soc Lond B Biol Sci 356: 12931322.
40. Sled JG, Zijdenbos AP, Evans AC (1998) A nonparametric method for
automatic correction of intensity nonuniformity in MRI data. IEEE Transactions on Medical Imaging 17: 8797.
143155.
42. Zijdenbos AP, Forghani R, Evans AC (1998) Automatic quantification of MS
lesions in 3D MRI brain data sets: Validation of INSECT. In: III WMW,
Colchester ACF, Delp SL, eds. MICCAI Springer, volume 1496 of Lecture
Notes in Computer Science. pp 439448.
43. Tohka J, Zijdenbos A, Evans A (2004) Fast and robust parameter estimation for
statistical partial volume models in brain mri. NeuroImage 23: 8497.
44. Kim JS, Singh V, Lee JK, Lerch J, Ad-Dabbagh Y, et al. (2005) Automated 3-d
extraction and evaluation of the inner and outer cortical surfaces using a
laplacian map and partial volume effect classification. NeuroImage 27: 210221.
45. Chung M, Taylor J (2004) Diffusion smoothing on brain surface via finite
element method. In: ISBI IEEE. pp 432435.
1. Brookmeyer R, Johnson E, Ziegler-Graham K, Arrighi HM (2007) Forecasting

the global burden of Alzheimers disease. Alzheimers and Dementia 3: 186191.
2. Cummings J, Doody R, Clark C (2007) Disease-modifying therapies for
alzheimer disease: challenges to early intervention. Neurology 69: 16221634.
3. Petersen R (2001) Practice parameter: early detection of dementia: mild
cognitive impairment (an evidence-based review). Neurology 56: 1133.
4. Gauthier S, Reisberg B, Zaudig M, Petersen R, Ritchie K, et al. (2006) Mild
cognitive impairment. Lancet 367: 12621270.
5. Braak H, Braak E (1991) Neuropathological stageing of Alzheimer-related
changes. Acta Neuropathologica 82: 239259.
6. Wenk G (2003) Neuropathologic changes in alzheimers disease. Journal of
Clinical Psychiatry 64 Suppl 9.
7. Hampel H, Brger K, Teipel SJ, Bokde AL, Zetterberg H, et al. (2008) Core
candidate neurochemical and imaging biomarkers of alzheimers disease.
Alzheimers and Dementia 4: 3848.
8. Hyman BT, Marzloff K, Arriagada PV (1993) The lack of accumulation of senile
plaques or amyloid burden in Alzheimers disease suggests a dynamic balance
between amyloid deposition and resolution. J Neuropathol Exp Neurol 52:
594600.
9. Gmez-Isla T, Hollister R, West H, Mui S, Growdon J, et al. (1997) Neuronal loss
correlates with but exceeds neurofibrillary tangles in alzheimers disease. Ann
Neurol 41: 1724.
10. Ingelsson M, Fukumoto H, Newell KL, Growdon JH, Hedley-Whyte ET, et al.
(2004) Early abeta accumulation and progressive synaptic loss, gliosis, and tangle
formation in ad brain. Neurology 62: 92531.
11. Jack C, Jr., Knopman D, Jagust W, Shaw L, Aisen P, et al. (2010) Hypothetical
model of dynamic biomarkers of the alzheimers pathological cascade. Lancet
12. Hampel H, Frank R, Broich K, Teipel S, Katz J, abd Hardy RG, et al. (2010)
Biomarkers for alzheimers disease: academic, industry and regulatory
perspectives. NatRevDrug Discov 9: 560574.
13. Fan Y, Resnick SM, Wu X, Davatzikos C (2008) Structural and functional
biomarkers of prodromal alzheimers disease: A high-dimensional pattern
classification study. NeuroImage 41: 277285.
14. Hinrichs C, Singh V, Xu G, Johnson S (2009) Mkl for robust multi-modality ad
classification. In: MICCAI (II) Springer, volume 5762 of Lecture Notes in
Computer Science. pp 786794.
15. Davatzikos C, Bhatt P, Shaw LM, Batmanghelich KN, Trojanowski JQ (2010)
Prediction of MCI to AD conversion, via MRI, CSF biomarkers, and pattern
classification. Neurobiology of Aging In Press, Corrected Proof: -.
16. Eckerstrom C, Andreasson U, Olsson E, Rolstad S, Blennow K, et al. (2010)
Combination of hippocampal volume and cerebrospinal uid biomarkers
improves predictive value in mild cognitive impairment. Dement Geriatr Cogn
Disord 29: 294300.
17. Kohannim O, Hua X, Hibar DP, Lee S, Chou YY, et al. (2010) Boosting power
for clinical trials using classifiers based on multiple biomarkers. Neurobiology of
Aging 31: 14291442.
18. Landau S, Harvey D, Madison C, Reiman E, Foster N, et al. (2010) Comparing
predictors of conversion and decline in mild cognitive impairment. Neurology
75: 2308.
19. Walhovd KB, Fjell AM, Brewer J, McEvoy LK, Fennema-Notestine C, et al.
(2010) Combining MR Imaging, Positron-Emission Tomography, and CSF
Biomarkers in the Diagnosis and Prognosis of Alzheimer Disease. American
Journal of Neuroradiology 31: 347354+.
20. Wolz R, Aljabar P, Hajnal JV, Lotjonen J, Rueckert D (2011) Manifold learning
combining imaging with non-imaging information. In: IEEE International
Symposium on Biomedical Imaging. pp 960963.
21. Chupin M, Hammers A, Liu R, Colliot O, Burdett J, et al. (2009) Automatic
segmentation of the hippocampus and the amygdala driven by hybrid
constraints: Method and validation. NeuroImage 46: 749761.
22. Devanand DP, Pradhaban G, Liu X, Khandji A, De Santi S, et al. (2007)
Hippocampal and entorhinal atrophy in mild cognitive impairment - Prediction
of Alzheimer disease. Neurology 68: 828836+.
23. Gerardin E, Chetelat G, Chupin M, Cuingnet R, Desgranges B, et al. (2009)
Multidimensional classification of hippocampal shape features discriminates
Alzheimers disease and mild cognitive impairment from normal aging.
37
46. Brun CC, Lepor N, Pennec X, Lee AD, Barysheva M, et al. (2009) Mapping the
regional inuence of genetics on brain structure variability a tensor-based
morphometry study. NeuroImage 48: 3749.
47. Koikkalainen J, Lotjonen J, Thurfjell L, Rueckert D, Waldemar G, et al. (2011)
Multi-template tensor-based morphometry: Application to analysis of alzheimers disease. NeuroImage 56: 11341144.
48. Belkin M, Niyogi P (2003) Laplacian eigenmaps for dimensionality reduction
and data representation. Neural Computation 15: 13731396.
49. Rueckert D, Sonoda LI, Hayes C, Hill DLG, Leach MO, et al. (1999) Nonrigid
registration using free-form deformations: Application to breast MR images.
IEEE Transactions on Medical Imaging 18: 712721.
50. Wolz R, Heckemann RA, Aljabar P, Hajnal JV, Hammers A, et al. (2010)
Measurement of hippocampal atrophy using 4D graph-cut segmentation:
Application to ADNI. NeuroImage 52: 10091018.
51. Draper NR, Smith H (1998) Applied Regression Analysis (Wiley Series in
Probability and Statistics). Wiley.
52. Krzanowski WJ (1988) Principles of Multivariate Analysis: A Users Perspective
Oxford University Press.
53. Cortes C, Vapnik V (1995) Support-vector networks. Machine Learning 20:
273297.
54. Vemuri P, Gunter JL, Senjem ML, Whitwell JL, Kantarci K, et al. (2008)
Alzheimers disease diagnosis in individual subjects using structural mr images:
Validation studies. NeuroImage 39: 11861197.
55. Westman E, Simmons A, Zhang Y, Muehlboeck JS, Tunnard C, et al. (2011)
Multivariate analysis of mri data for alzheimers disease, mild cognitive
impairment and healthy controls. NeuroImage 54: 11781187.
38
CSF and Brain Structural Imaging Markers of the

Alzheimers Pathological Cascade
Xianfeng Yang1, Ming Zhen Tan1, Anqi Qiu1,2,3*
1 Department of Bioengineering, National University of Singapore, Singapore, Singapore, 2 Singapore Institute for Clinical Sciences, the Agency for Science, Technology
and Research, Singapore, Singapore, 3 Clinical Imaging Research Center, National University of Singapore, Singapore, Singapore
Abstract
Cerebral spinal fluid (CSF) and structural imaging markers are suggested as biomarkers amended to existing diagnostic
criteria of mild cognitive impairment (MCI) and Alzheimers disease (AD). But there is no clear instruction on which markers
should be used at which stage of dementia. This study aimed to first investigate associations of the CSF markers as well as
volumes and shapes of the hippocampus and lateral ventricles with MCI and AD at the baseline and secondly apply these
baseline markers to predict MCI conversion in a two-year time using the Alzheimers Disease Neuroimaging Initiative (ADNI)
cohort. Our results suggested that the CSF markers, including Ab42, t-tau, and p-tau, distinguished MCI or AD from NC,
while the Ab42 CSF marker contributed to the differentiation between MCI and AD. The hippocampal shapes performed
better than the hippocampal volumes in classifying NC and MCI, NC and AD, as well as MCI and AD. Interestingly, the
ventricular volumes were better than the ventricular shapes to distinguish MCI or AD from NC, while the ventricular shapes
showed better accuracy than the ventricular volumes in classifying MCI and AD. As the CSF markers and the structural
markers are complementary, the combination of them showed great improvements in the classification accuracies of MCI
and AD. Moreover, the combination of these markers showed high sensitivity but low specificity for predicting conversion
from MCI to AD in two years. Hence, it is feasible to employ a cross-sectional sample to investigate dynamic associations of
the CSF and imaging markers with MCI and AD and to predict future MCI conversion. In particular, the volumetric
information may be good for the early stage of AD, while morphological shapes should be considered as markers in the
prediction of MCI conversion to AD together with the CSF markers.
Citation: Yang X, Tan MZ, Qiu A (2012) CSF and Brain Structural Imaging Markers of the Alzheimers Pathological Cascade. PLoS ONE 7(12): e47406. doi:10.1371/
journal.pone.0047406
Editor: Yong Fan, Institution of Automation, CAS, China
Received April 13, 2012; Accepted September 13, 2012; Published December 19, 2012
Copyright: 2012 Yang et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
Funding: The work was supported by grants: a centre grant from the National Medical Research Council (NMRC/CG/NUHS/2010), the Young Investigator Award
at the National University of Singapore (NUSYIA FY10 P07), and the National University of Singapore MOE AcRF Tier 1. The funders had no role in study design,
data collection and analysis, decision to publish, or preparation of the manuscript.
Introduction
been suggested as improved predictors for MCI conversion to AD.

For instance, changes of hippocampal shapes between baseline
and a 2-year follow-up predicted MCI-AD conversion up to 80%
accuracy [11,12]. Unfortunately, to date, no clear, authoritative
instruction on which structural MRI measures are to be associated
with MCI and AD is available. This could be, partly, a result of the
extensive variety of image analysis techniques available. In
addition, while recent studies [13,14] have tested the feasibility
of baseline structural volumes and CSF in predicting conversion
from MCI to AD, performance of baseline structural shapes with
CSF markers for MCI-AD conversion remains relatively unknown.
In this paper, we first evaluated the hypothetical model
suggested by Jack et al. [7] through a cross-sectional study on
the Alzheimers Disease Neuroimaging Initiative (ADNI) cohort.
For this, we employed a supervised, multivariate classification
method, support vector machine (SVM), to distinguish MCI and
AD from normal aging. Features used include CSF biomarkers
and the shapes and volumes of hippocampus and lateral ventricles.
The hippocampi and lateral ventricles were chosen for their wellvalidated
status
as
prominent
hallmarks
of
AD
[8,15,16,17,18,19,20]. Subsequently, we aim to predict MCI
conversion to AD over a two-year follow-up period using baseline
Cerebral spinal fluid (CSF) and imaging markers have been

suggested as biomarkers to augment existing diagnostic criteria of
both mild cognitive impairment (MCI) and Alzheimers disease
(AD) [1,2,3,4,5,6]. Jack et al. [7] proposed a possible hypothetical
model in which biomarkers were temporally arranged in order of
abnormality along the pathological cascade of AD. In this model,
abnormal CSF Ab42 could occur two decades before the first
dementia-related symptoms, reaching a plateau prior any manifestation of cognitive impairment. In comparison to trajectories of
CSF tau, magnetic resonance imaging (MRI) markers surfaced
much later and were well correlated with the severity of AD
symptoms. However, this hypothetical model was directly derived
from longitudinal studies, where statistical inferences were founded
primarily on the rate of change of AD-related biomarkers over
time. For instance, increased rates of ventricular expansion and
brain atrophy in the medial temporal lobe were found to be
significantly correlated with cognitive decline, with good predictions for MCI to AD conversion [8] [9] [10].
Beyond simple volumetric measures, morphological shape of the
brain captures not only the degree of tissue loss but also its precise
anatomical location. As such, brain shape measures have since
December 2012 | Volume 7 | Issue 12 | e47406
39
CSF and Brain Shapes in MCI and AD
Structural Delineation. We automatically delineated the

hippocampus (HC) and lateral ventricles (LV) from the intensity
inhomogeneity corrected T1-weighted MR images using FreeSurfer [23]. Due to the lack of constraints on structural shapes, this
process introduced irregularities and topological errors (e.g. holes)
at the hippocampal and ventricular boundary. This would increase
shape variation and thus reduces statistical power to detect group
differences. To avoid this pitfall, we generated the hippocampal or
ventricular shape of each individual subject with the properties of
smoothness and correct topology by injecting an atlas shape into
them using the large deformation diffeomorphic metric image
mapping algorithm [24]. The hippocampal and lateral ventricular
atlas shapes were created from 41 manually labeled hippocampi
and lateral ventricles via a large deformation diffeomorphic atlas
generation algorithm [25]. Each hippocampal (or lateral ventricular) volume was approximated by the transformed atlas through
the LDDMM transformation. The reader is referred to [26] for
the mathematical derivation of this atlas injection procedure and
its evaluation as well as the segmentation accuracy on the
hippocampus and lateral ventricles. This delineation approach has
been successfully applied to investigate the hippocampus and other
subcortical shapes in AD [22].
We constructed the surface representation of the hippocampus
and lateral ventricles by composing the LDDMM transformation
on the corresponding atlas surfaces [27]. The left and right
hippocampal surfaces were respectively constructed using 2364
triangles with 1184 vertices and 2458 triangles with 1231 vertices,
while the left and right lateral ventricular surfaces were
respectively composed of 6966 triangles with 3485 vertices and
7890 triangles with 3947 vertices. The average triangle area and
edge length of the hippocampal surfaces were respectively
0.59 mm2 and 1.2 mm, while those of the ventricular surfaces
were respectively 0.64 mm2 and 1.1 mm. Hence, the size of the
triangles was comparable to the image resolution (1 mm3).
ISOMAP Shape Embedding. Unlike the scalar volume
measure, structural shapes lie on a high dimensional space, which
makes it challenging for statistical inference. In this study, we
employed ISOMAP [28] to embed the shapes of the hippocampus
and lateral ventricles into a Euclidean space with a few dimensions
such that this low-dimensional embedding is quasi-isometric to the
shapes in the high dimensional space. For this, we first computed
diffeomorphic metric distances between any two shapes using their
first order approximation described in [29] and constructed a pairwise distance matrix. ISOMAP then found a Euclidean lowdimensional representation of the shapes that preserved the
relationship of any two shapes described in the pair-wise distance
matrix. These Euclidean coordinates were obtained by finding
eigenvectors corresponding to the largest eigenvalues of the kernel
matrix stemmed from the distance matrix reshaped by a centering
matrix [28]. The dimension of the eigenvectors is the same as the
number of subjects, and each eigenvector is one component or one
dimension of the ISOMAP shape embedding. Using this approach
with all 383 subjects, the bilateral hippocampal shapes can be
characterized using the first 20 ISOMAP components whose
Euclidean distance matrix is highly correlated with the pair-wise
metric distance matrix generated using the first order approximation of the diffeomorphic metric (Pearsons Correlation: r = 0.91).
The bilateral lateral ventricular shapes can be represented using
the first 20 ISOMAP components whose Euclidean distance
matrix is very much similar to the pair-wise metric distance matrix
(Pearsons Correlation: r = 0.97).
CSF and MRI measures. In particular, both volumetric and shape

analyses were applied to compare their sensitivity and specificity in
the prediction of MCI conversion to AD.
Methods
The ADNI was launched in 2003 by the National Institute on
Aging (NIA), the National Institute of Biomedical Imaging and
Bioengineering (NIBIB), the Food and Drug Administration [21],
private pharmaceutical companies and non-profit organizations,
as a $60 million, 5-year publicprivate partnership. The primary
goal of ADNI has been to test whether serial MRI, PET, other
biological markers, and clinical and neuropsychological assessments can be combined to measure the progression of MCI and
early AD. Determination of sensitive and specific markers of very
early AD progression is intended to aid researchers and clinicians
to develop new treatments and monitor their effectiveness, as well
as lessen the time and cost of clinical trials.
ADNI is the result of efforts of many coinvestigators from a
broad range of academic institutions and private corporations, and
subjects have been recruited from over 50 sites across the U.S. and
Canada. The initial goal of ADNI was to recruit 800 adults, ages
55 to 90, to participate in the research approximately 200
cognitively normal older individuals to be followed for 3 years, 400
people with MCI to be followed for 3 years, and 200 people with
early AD to be followed for 2years (see www.adni-info.org for upto-date information). The data were analyzed anonymously, using
publicly available secondary data from the ADNI study, therefore
no ethics statement is required for this work.
Subjects
The ADNI general eligibility criteria are described at www.
adni- info.org. Briefly, subjects are between 5590 years of age,
having a study partner able to provide an independent evaluation
of functioning. Specific psychoactive medications will be excluded.
General inclusion/exclusion criteria are as follows: 1) healthy
subjects: Mini- Mental State Examination (MMSE) scores between
2430, a Clinical Dementia Rating (CDR) of 0, non-depressed,
non-MCI, and nondemented; 2) MCI subjects: MMSE scores
between 2430, a memory complaint, having objective memory
loss measured by education adjusted scores on Wechsler Memory
Scale Logical Memory II, a CDR of 0.5, absence of significant
levels of impairment in other cognitive domains, essentially
preserved activities of daily living, and an absence of dementia;
and 3) mild AD: MMSE scores between 2026, CDR of 0.5 or 1.0,
and meets the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimers Disease and
Related Disorders Association (NINCDS/ADRDA) criteria for
probable AD.
In this study, 383 subjects were chosen from our previous study
[22]. Within this group, 218 have both MRI and CSF baseline
data (age: 74.467.2 years), with 72 normal controls (NC), 35 AD
subjects, and 111 MCI patients. Amongst these 383 subjects, 25
subjects with MCI converted to AD within 24 months.
Structural MR scans were collected across a variety of scanners
with protocols individualized for each scanner, as defined at www.
loni.ucla.edu/ADNI/Research/Cores/index.shtml. The CSF
Ab42, t-tau and p-tau data were downloaded from the ADNI
web site (www.loni.ucla.edu/ADNI).
MRI analysis
40
Figure 1 illustrates the MRI data processing that is detailed

below.

Figure 1. Schematic of MRI data processing.

Statistical Analysis
Markers at Stages of MCI an AD
A linear support vector machine (SVM) [30] was employed to

identify diagnosis of subjects from any two groups (NC, MCI, and
AD). The SVM classifier seeks the optimal decision boundary that
has a maximal margin closest to the training samples such that
generalization error bound can be minimized. Hence, the SVM
classifier is robust to outliers. In our study, we independently and
jointly considered volumes (or shapes) of the hippocampus, lateral
ventricles and CSF markers as the SVM input features to identify
subjects with MCI and AD from NC and MCI from AD. Shape
features comprise only of ISOMAP components with significant
group differences based on Student t-tests.
Our study had fewer AD subjects as compared to the NC or
MCI groups. To resolve any influences of unequal sample sizes
among the NC, MCI, and AD groups on the classification
accuracy, we employed random sampling to reduce the number of
NC and MCI subjects such that all three groups have equal sample
sizes. This was repeated for 100 times. For each trial, leave-oneout cross-validation was adopted to estimate the classification
accuracy. The confidence interval of the classification accuracy
was computed.
Moreover, we also applied the SVM to test the sensitivity and
specificity for predicting MCI conversion in a two-year time
window when volumes (or shapes) of the hippocampus and lateral
ventricles as well as CSF markers assessed at the baseline were
independently or jointly considered as features in the SVM.
Hippocampal volume and shape markers. Bilateral hippocampal volumes distinguished MCI and AD subjects from
normal controls at an accuracy of 61.9% and 65.5% respectively,
with relatively high specificity (MCI: 66.1%; AD: 73.3%) but low
sensitivity (MCI: 57.7%; AD: 57.8%) (Table 2). However, the
hippocampal volumes lost statistical power in the separation of
subjects with MCI and AD (classification accuracy: 42.3%;
sensitivity: 45.3%; specificity: 39.2%, Table 2).
In shape analysis, the first 20 ISOMAP components characterized bilateral hippocampal shapes among all 383 subjects. The 1st,
2nd, 3rd, 4th, 9th, 13th, 18th components contributed to hippocampal shape differences between NC and AD. Among these
components, most of them (the 1st, 2nd, 3rd, 7th, 13th) also
contributed to shape differences between NC and MCI (Table 3).
Interestingly, only the 4th and 9th components showed shape
differences between MCI and AD. Moreover, among these
ISOMAP components, the 1st and 2nd components were highly
correlated with the hippocampal volume (Pearson correlations:
r = 0.7, p,0.01 for the 1st component; r = 0.5, p,0.01 for the 2nd
component), which were dominant components for group
differences in hippocampal shapes between NC and MCI. This
is illustrated as relatively homogeneous shrinkage over the bilateral
hippocampi in Figure 2 (a,b). But the 4th and 9th components
were not associated with the hippocampal volume (p.0.05),
suggesting that only local hippocampal shapes contributed to the
difference between MCI and AD, as seen in Figure 2 (c,d). This
can be further supported by evidence of increased classification
accuracy rates for the classifications between NC and AD (79.2%),
NC and MCI (67.4%), and MCI and AD (57.2%) (Table 2) when
the ISOMAP embedding of hippocampal shapes were used in the
SVM classifiers.
Lateral ventricular volume and shape markers. The
volumes of bilateral lateral ventricles distinguished subjects with
MCI and AD from normal controls at the accuracy of 63.1% and
65.5% respectively, with relatively high specificity (MCI: 75.3%;
AD: 72.3%) but low sensitivity (MCI: 50.9%; AD: 58.8%)
(Table 2). However, volumes of the lateral ventricles lost statistical
Results
Demographic Information
Demographic information for different diagnosis groups at
baseline are shown in Table 1. No significant differences in age
were found among the NC, MCI, and AD groups (ANOVA,
p = 0.454). 25 out of 111 MCI subjects were diagnosed as AD at
the two-year follow up and were denoted as the MCI-c group.
Rest of the MCI subjects were placed in the MCI-s group. No
significant MMSE difference was found between the two MCI
groups at baseline (p = 0.10).
41
Table 1. Demographic information for each of the diagnosis groups (normal controls (NC), mild cognitive impairment (MCI), and
Alzheimers disease (AD)) at the baseline.
Group
Subjects n
Age (meanSD)
Gender (female/male)
MMSE (meanSD)
NC
72
75.265.2
35/37
2961
MCI-s
86
7467.7
28/58
26.761.8
MCI-c
25
73.566.9
6/19
26.161.5
AD
35
74.669.3
15/20
22.961.8
Note: SD standard deviation; MMSE mini-mental state examination; MCI-s subjects with MCI who remained as MCI at the two-year follow up; MCI-c subjects with
MCI who converted as AD at the two-year follow up.
power in separating subjects with MCI and AD (classification

accuracy: 30.4%; sensitivity: 21.2%; specificity: 39.7%, Table 2).
In shape analysis, the first 20 ISOMAP components characterized bilateral lateral ventricular shapes among all 383 subjects.
The 1st, 7th, 8th, 9th, 17th, 19th components contributed to
hippocampal shape differences between NC and AD. Among
these components, several of them (the 1st, 8th) also contributed to
shape differences between NC and MCI (Table 3). Interestingly,
only the 9th, 13th, and 17th components showed the shape
differences between MCI and AD. Moreover, among these
ISOMAP components, the 1st component was highly correlated
with the ventricular volume (Pearson correlations: r = 0.98,
p,0.01), though it was not the only component contributing to
shape differences between NC and MCI, as illustrated in Figure 3
(a,b). However, the 9th, 13th, and 17th components were not
associated with the lateral ventricular volume (p.0.05), suggesting
that only local ventricular shapes contributed to the difference
between MCI and AD. This can also be seen in Figure 3 (c,d).
Unlike the hippocampus, the lateral ventricular shapes did not
lead to better classification accuracy rates between NC and AD
(61.5%), and between NC and MCI (59%) when compared with

the lateral ventricular volumes (see Table 2). However, the
ventricular shapes achieved markedly better accuracy in distinguishing MCI and AD (60.1%) (Table 2).
CSF Markers. Students t-tests revealed that all three CSF
markers, Ab42, t-tau, and p-tau, showed statistically significant
differences between NC and AD and between NC and MCI
(Table 3). However, only Ab42 showed group differences
between MCI and AD (Table 3). The SVM classification
revealed that the CSF markers can distinguish NC and AD at
the classification accuracy of 81.4%, higher than those based on
the hippocampal or ventricular imaging markers (Table 2).
Additionally, the CSF markers achieved similar accuracy in
distinguishing NC and MCI, and MCI and AD subjects in
comparison with hippocampal and ventricular shapes. Interestingly, the CSF markers gave higher sensitivities than the imaging
markers (Table 2).
The
combination
of
the
Imaging
and
CSF
Markers. Combining the CSF markers and the volumes of
the hippocampus and lateral ventricles as features in the SVM
Table 2. The classification accuracy, sensitivity, and specificity of the support vector machine (SVM) classifiers are given for
distinguishing normal controls (NC) and subjects with Alzheimers disease (AD), NC and subjects with mild cognitive impairment
(MCI), and subjects with MCI and AD.
NC.vs. AD
NC.vs. MCI
MCI.vs. AD
Hippocampal Volumes and Shapes

Hp volumes
65.5% (CI: 64.4%,66.6%)

Sensitivity = 57.8%, Specificity = 73.3%
61.9% (CI: 61%,62.8%)

42.3% (CI:32.2%,52.3%)
Hp shapes
79.2% (CI: 78.5%,80%)

67.4% (CI: 66.2%,68.6%)

Sensitivity = 64%, Specificity = 70.8%
57.2% (CI: 51.7%,62.8%)

Lateral Ventricular Volumes and Shapes

LV volumes
65.5% (CI: 64.9%,66.1%)

Sensitivity = 58.8% Specificity = 72.3%
63.1% (CI: 62%,64.2%)

30.4% (CI: 18.8%,42%)

Sensitivity = 21.2% Specificity = 39.7%
LV shapes
61.5% (CI: 59.2%,63.9%)

59% (CI: 56.1%,62%)

60.1% (CI:57.1%,63.1%)
Sensitivity = 62%, Specificity = 58.3%
81.4% (CI: 80.3%,82.5%)

68.4% (CI: 67.8%,69%)

61.3% (CI: 59.4%,63.2%)

Sensitivity = 84.3%, specificity = 38.3%
CSF, Hp volumes,
LV volumes
85.4% (CI: 84.3%,86.5%)

Sensitivity = 88.8%, Specificity = 82%
72% (CI: 70.5%,73.5%)

60.9% (CI: 59.1%,62.8%)

CSF, Hp shapes,
LV shapes
92.2% (CI: 91%,93.5%)

70.3% (CI: 68.6%,72.9%)

Sensitivity = 69.5%, specificity = 71.9%
69.6% (CI: 66.4%,72.8%)

CSF
CSF markers
Combination
42
The volumes and shapes of the hippocampus (Hp) and lateral ventricles (LV) as well as cerebral spinal fluid (CSF) markers are respectively used as features in the SVM.

Prediction of the MCI Conversion

To predict MCI conversion at a two-year follow-up, we trained
an NC and AD classifier using the CSF and imaging markers
before applying it to the 25 MCI converters and 86 MCI nonconverters. Again, combination of the CSF markers with the
hippocampus and lateral ventricles shapes at baseline showed the
best prediction (66.7%) and sensitivity (82%) for identifying the
MCI converters when compared to individual imaging or CSF
markers (Table 4).
Discussion
Our study demonstrated the dynamic trajectories of the CSF,
hippocampal and lateral ventricular markers in the Alzheimers
pathological cascade using a cross-sectional ADNI sample and also
showed the feasibility of predicting future MCI-to-AD conversion
using baseline CSF and imaging markers. The CSF markers,
including Ab42, t-tau, and p-tau, distinguished MCI or AD from
NC, while only the Ab42 CSF marker contributed to the
differentiation between MCI and AD. The hippocampal shapes
performed better than the hippocampal volumes in classifying NC
and MCI, NC and AD, as well as MCI and AD. Interestingly, as
compared to the ventricular shape, ventricular volume performed
better in distinguishing MCI or AD from NC. The ventricular
shape, however, showed better accuracy in the classification for
MCI and AD. As the CSF and structural markers were
complementary, their combination showed great improvement
in the classification accuracies at all the stages of AD. Moreover,
the combination of these baseline markers also showed high
sensitivity but low specificity for predicting MCI conversion to AD
during a two-year period.
Our findings supported the conclusion drawn in previous
studies [31]; [32], where abnormality of both CSF Ab42 and
neurodegenerative biomarkers, including CSF tau and MRI
markers, precedes clinical symptoms; all these markers showed
significant differences between NC and MCI groups. However,
our findings did not support the hypothesis where CSF Ab42
reaches a plateau before the appearance of MRI atrophy and
cognitive symptoms, and remain static thereafter [7]. In our study,
CSF Ab42 continued to show appreciable power discriminant
between MCI and AD. In contrast, CSF tau lost its discriminating
power in distinguishing MCI and AD patients, suggesting that
CSF Ab42 reaches its plateau after CSF tau in the Alzheimers
pathological cascade.
MCI and AD patients were not well-separated using hippocampal volume, implying that the overall tissue loss in the
hippocampus may not be a good marker for monitoring AD
progression. However, we may not conclude that the hippocampus
Figure 2. Hippocampal shape differences among normal

controls (NC), mild cognitive impairment (MCI), and Alzheimers Disease (AD). Panels (a,b) respectively show group differences
in the left and right hippocampal surface deformations between MCI
and NC. Panels (c,d) respectively show group differences in the left and
right hippocampal surface deformations between AD and MCI. Warm
color denotes regions where structures have surface outward-deformation in the former group when compared with the latter group,
while cool color denotes regions where structures have surface inwarddeformation in the former group when compared with the latter group.
increased the classification accuracies between NC and AD

(85.4%) and between NC and MCI (72%) when compared to
those achieved using only one type of markers (see Table 2).
Nevertheless, there was no improvement on the classification
accuracy between MCI and AD when compared with that using
the CSF markers alone.
Combining the CSF markers and the shapes of the hippocampus and lateral ventricles as features in the SVM increased the
classification accuracies between NC and AD (92.2%), between
NC and MCI (70.3%), and between MCI and AD (69.6%) when
compared to those achieved using only one type of the markers
(see Table 2).
Shapes of the two structures with the CSF markers performed
better in separating NC and MCI (p,0.001) or NC and AD
(p,0.001) when compared to features combining the volumes and
CSF markers. However, shapes of the two structures with the CSF
markers performed worse in separating NC and MCI when
compared to features combining the volumes and CSF markers
(p = 0.013).
Table 3. ISOMAP components of the hippocampus and lateral ventricles as well as CSF markers contribute to the group
differences between normal controls (NC) and subjects with Alzheimers disease (AD), NC and subjects with mild cognitive
impairment (MCI), and subjects with MCI and AD.
NC.vs. AD
NC.vs. MCI
MCI vs AD
Imaging Markers
ISOMAP components
ISOMAP components
ISOMAP components
hippocampal shapes
1,2,3,4,9,13,18
1,2,3,7,13
4,9
lateral ventricular shapes
1,7,8,9,17,19
1,8,12,14
9, 13,17
Ab42, t-tau, p-tau
Ab42, t-tau, p-tau
Ab42
CSF Markers
CSF Ab42, t-tau, p-tau
43
5
Figure 3. Shape differences of the lateral ventricles among normal controls (NC), mild cognitive impairment (MCI), and Alzheimers
Disease (AD). Panels (a,b) respectively show group differences in the left and right lateral ventricular surface deformations between MCI and NC.
Panels (c,d) respectively show group differences in the left and right lateral ventricular surface deformations between AD and MCI. Warm color
denotes regions where structures have surface outward-deformation in the former group when compared with the latter group, while cool color
denotes regions where structures have surface inward-deformation in the former group when compared with the latter group.
reached its abnormality peak before the late stage of AD, as its
local shape variations were significantly associated with progression from MCI to AD. Our results also showed that such local
shape markers aided the hippocampal markers in achieving
slightly better accuracy than the CSF markers in the prediction for
MCI conversion to AD. This was also supported by previous
studies, suggesting that MRI markers (e.g. cortical thickness of the
medial temporal lobe) correlate well with severity of cognitive
impairment and have greater predictive power than the CSF tau
[33]. Based on these evidences, we may conclude that the
hippocampal shape marker reaches its plateau after CSF tau.
However, the order in which the CSF Ab42 and MRI markers
reach their abnormality peaks is still unclear based on our current
study.
The volume of the lateral ventricles cannot distinguish MCI and
AD patients. Interestingly, the overall expansion of the lateral
ventricles showed better performance in identifying MCI or AD
from NC when compared with their shapes. This result agrees
with previous studies, suggesting that rates of ventricular
expansion were significantly different between AD (or MCI) and
NC groups [8]. This implies that complicated shape analysis might
not always be necessary to provide better structural morphological
markers when compared to volumetric analysis. Even for the same

structure, its markers can be different at different stages of AD.
Again, we may not conclude that the lateral ventricles reached
their abnormality peak before the late stage of AD, as their local
shape variations were significantly associated with progression
from MCI to AD. Likewise, ventricular shape markers slightly
outperformed CSF markers in the prediction for MCI conversion
to AD, as verified by previous studies [34], wherein clearer
correlation was observed for ventricular volumes against worsening cognitive indices, as compared to CSF biomarkers.
Our study showed that the CSF and structural markers are
complementary to each other in the AD pathological cascade.
This suggests that the CSF markers, (Ab42, t-tau, and p-tau) with
the volumes of the hippocampus and lateral ventricles, is a good
combination for distinguishing NC and MCI, while CSF Ab42
marker with the shape of the hippocampus and lateral ventricles is
a good combination for identifying MCI and AD.
As the shapes of the hippocampus and lateral ventricles
contributed more to the difference between MCI and AD than
their volumes, our study further showed that the combination of
the CSF markers and the shapes of the two structures at baseline
predicted MCI conversion to AD in the two-year follow up at an
Table 4. The accuracy, sensitivity, and specificity for predicting the MCI converters are listed when the volumes or shapes of the
hippocampus (Hp) or the lateral ventricles (LV), or the CSF markers, or their combination were used as features in the classification.
Markers
Accuracy (95% CI)
Sensitivity
Specificity
Hippocampal volumes and Shapes

Hp volumes
54%(53.1%,54.9%)
54.4%
53.6%
Hp shapes
63%(62.1%,64%)
74.9%
51.2%
Lateral Ventricular Volumes and Shapes

LV volumes
55.6%(54.7%,56.5)
63.7%
47.5%
LV shapes
62.7%(61.5%,63.9%)
67.9%
57.5%
62.2%(61.3%,63.1%)
80.4%
44%
CSF, Hp volumes, LV volumes 59.2%(58.3%,60%)
81.1%(80.5%,81.6%)
37.2%(35.6%,38.9%)
CSF, Hp shapes, LV shapes
82%(81.5%,82.6%)
51.4% (49.4%,53.3%)
CSF
CSF markers
Combination
44
66.7%(65.7%,67.8%)

improved accuracy of 66.7%. Previous studies [35] achieved

similar prediction accuracy (68.5%) for the MCI conversion within
a three-year follow up, with suggestions claiming that multiple
predictors, including the CSF markers, hippocampal volume,
entorhinal cortex thickness, etc. would not perform better than a
single predictor. This differs from the conclusion derived from our
findings, possibly because structural shape measures contain more
complementing features than structural volume measures. The
combination of the shapes and CSF markers achieved high
classification accuracy (92.2%) between NC and AD, improving
by more than 10% over the CSF markers. This result is
comparable to those previously reported where CSF, MRI and
PET imaging markers were combined [36].
In summary, we conclude that it is feasible to employ a crosssectional sample to investigate dynamic associations of the CSF
and imaging markers with MCI and AD and to predict future
MCI conversion to AD. In particular, volumetric information may
be good for the early stages of AD while morphological shapes
should be considered as markers in the prediction of MCI

conversion to AD together with the CSF markers.
Acknowledgments
Data used in preparation of this article were obtained from the Alzheimers
Disease Neuroimaging Initiative (ADNI) database (adni.loni.ucla.edu). As
such, the investigators within the ADNI contributed to the design and
implementation of ADNI and/or provided data but did not participate in
analysis or writing of this report. A complete listing of ADNI investigators
can be found at: http://adni.loni.ucla.edu/wp-content/uploads/how_to_
apply/ADNI_Acknowledgement_List.pdf.
Conceived and designed the experiments: XY MZT AQ. Performed the
experiments: XY MZT AQ. Analyzed the data: XY MZT AQ.
Contributed reagents/materials/analysis tools: XY MZT AQ. Wrote the
paper: XY MZT AQ.
References
1. Jack CR, Jr., Albert MS, Knopman DS, McKhann GM, Sperling RA, et al.
(2011) Introduction to the recommendations from the National Institute on
Aging-Alzheimers Association workgroups on diagnostic guidelines for Alzheimers disease. Alzheimers Dement 7: 257262.
2. McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR, Jr., et al.
(2011) The diagnosis of dementia due to Alzheimers disease: recommendations
from the National Institute on Aging-Alzheimers Association workgroups on
diagnostic guidelines for Alzheimers disease. Alzheimers Dement 7: 263269.
3. Albert MS, DeKosky ST, Dickson D, Dubois B, Feldman HH, et al. (2011) The
diagnosis of mild cognitive impairment due to Alzheimers disease: recommendations from the National Institute on Aging-Alzheimers Association workgroups on diagnostic guidelines for Alzheimers disease. Alzheimers Dement 7:
270279.
4. Sperling RA, Aisen PS, Beckett LA, Bennett DA, Craft S, et al. (2011) Toward
defining the preclinical stages of Alzheimers disease: recommendations from the
National Institute on Aging-Alzheimers Association workgroups on diagnostic
guidelines for Alzheimers disease. Alzheimers Dement 7: 280292.
5. Ferrarini L, Palm WM, Olofsen H, van der Landen R, van Buchem MA, et al. (2008)
Ventricular shape biomarkers for Alzheimers disease in clinical MR images.
Magnetic resonance in medicine : official journal of the Society of Magnetic
Resonance in Medicine/Society of Magnetic Resonance in Medicine 59: 260267.
6. Ferrarini L, Frisoni GB, Pievani M, Reiber JH, Ganzola R, et al. (2009)
Morphological hippocampal markers for automated detection of Alzheimers
disease and mild cognitive impairment converters in magnetic resonance images.
Journal of Alzheimers disease : JAD 17: 643659.
7. Jack CR, Jr., Knopman DS, Jagust WJ, Shaw LM, Aisen PS, et al. (2010)
Hypothetical model of dynamic biomarkers of the Alzheimers pathological
cascade. Lancet Neurol 9: 119128.
8. Jack CR, Jr., Lowe VJ, Weigand SD, Wiste HJ, Senjem ML, et al. (2009) Serial
PIB and MRI in normal, mild cognitive impairment and Alzheimers disease:
implications for sequence of pathological events in Alzheimers disease. Brain
132: 13551365.
classification. Neurobiol Aging 32: 2322 e23192327.
10. Kantarci K, Weigand SD, Przybelski SA, Shiung MM, Whitwell JL, et al. (2009)
Risk of dementia in MCI: combined effect of cerebrovascular disease, volumetric
MRI, and 1H MRS. Neurology 72: 15191525.
11. Qiu A, Younes L, Miller MI, Csernansky JG (2008) Parallel transport in
diffeomorphisms distinguishes the time-dependent pattern of hippocampal
surface deformation due to healthy aging and the dementia of the Alzheimers
type. Neuroimage 40: 6876.
12. Costafreda SG, Dinov ID, Tu Z, Shi Y, Liu CY, et al. (2011) Automated
hippocampal shape analysis predicts the onset of dementia in mild cognitive
impairment. Neuroimage 56: 212219.
13. Westman E, Muehlboeck JS, Simmons A (2012) Combining MRI and CSF
measures for classification of Alzheimers disease and prediction of mild
cognitive impairment conversion. Neuroimage.
14. Cui Y, Liu B, Luo S, Zhen X, Fan M, et al. (2011) Identification of conversion
from mild cognitive impairment to Alzheimers disease using multivariate
predictors. PLoS One 6: e21896.
15. Cardenas VA, Du AT, Hardin D, Ezekiel F, Weber P, et al. (2003) Comparison
of methods for measuring longitudinal brain change in cognitive impairment
and dementia. Neurobiol Aging 24: 537544.
16. Ridha BH, Barnes J, Bartlett JW, Godbolt A, Pepple T, et al. (2006) Tracking
atrophy progression in familial Alzheimers disease: a serial MRI study. Lancet
Neurol 5: 828834.
17. Apostolova LG, Dutton RA, Dinov ID, Hayashi KM, Toga AW, et al. (2006)
Conversion of mild cognitive impairment to Alzheimer disease predicted by
hippocampal atrophy maps. Arch Neurol 63: 693699.
18. Fox NC, Warrington EK, Freeborough PA, Hartikainen P, Kennedy AM, et al.
(1996) Presymptomatic hippocampal atrophy in Alzheimers disease. A
longitudinal MRI study. Brain 119 (Pt 6): 20012007.
19. Wang L, Swank JS, Glick IE, Gado MH, Miller MI, et al. (2003) Changes in
hippocampal volume and shape across time distinguish dementia of the
Alzheimer type from healthy aging. Neuroimage 20: 667682.
20. Apostolova LG, Dinov ID, Dutton RA, Hayashi KM, Toga AW, et al. (2006) 3D
comparison of hippocampal atrophy in amnestic mild cognitive impairment and
Alzheimers disease. Brain 129: 28672873.
21. Alves Fda S, Figee M, Vamelsvoort T, Veltman D, de Haan L (2008) The revised
dopamine hypothesis of schizophrenia: evidence from pharmacological MRI studies
with atypical antipsychotic medication. Psychopharmacol Bull 41: 121132.
22. Qiu A, Fennema-Notestine C, Dale AM, Miller MI, Initiative AsDN (2009)
Regional shape abnormalities in mild cognitive impairment and Alzheimers
disease. Neuroimage 45: 656661.
23. Fischl B, Salat DH, Busa E, Albert M, Dieterich M, et al. (2002) Whole Brain
Segmentation: Automated Labeling of Neuroanatomical Structures in the
Human Brain. Neuron 33: 341355.
24. Miller MI, Qiu A (2009) The emerging discipline of Computational Functional
Anatomy. Neuroimage 45: S1639.
25. Qiu A, Brown T, Fischl B, Ma J, Miller MI (2010) Atlas Generation for
Subcortical and Ventricular Structures with Its Applications in Shape Analysis.
IEEE Transactions on Image Processing 19: 15391547.
26. Qiu A, Miller MI (2008) Multi-structure network shape analysis via normal
surface momentum maps. Neuroimage 42: 14301438.
27. Qiu A, Brown T, Fischl B, Ma J, Miller MI (2010) Atlas generation for
subcortical and ventricular structures with its applications in shape analysis.
IEEE Trans Image Process 19: 15391547.
28. Tenenbaum JB, de Silva V, Langford JC (2000) A global geometric framework
for nonlinear dimensionality reduction. Science 290: 23192323.
29. Yang X, Goh A, Qiu A (2011) Locally Linear Diffeomorphic Metric Embedding
(LLDME) for surface-based anatomical shape modeling. Neuroimage 56: 149
161.
30. Vapnik VN (1995) The Nature of Statistical Learning Theory: Springer.
31. Chetelat G, Desgranges B, De La Sayette V, Viader F, Eustache F, et al. (2002)
Mapping gray matter loss with voxel-based morphometry in mild cognitive
impairment. Neuroreport 13: 19391943.
32. Shaw LM, Vanderstichele H, Knapik-Czajka M, Clark CM, Aisen PS, et al.
(2009) Cerebrospinal fluid biomarker signature in Alzheimers disease neuroimaging initiative subjects. Ann Neurol 65: 403413.
33. Vemuri P, Wiste HJ, Weigand SD, Shaw LM, Trojanowski JQ, et al. (2009)
MRI and CSF biomarkers in normal, MCI, and AD subjects: predicting future
clinical change. Neurology 73: 294301.
34. Vemuri P, Wiste HJ, Weigand SD, Knopman DS, Trojanowski JQ, et al. (2010)
Serial MRI and CSF biomarkers in normal aging, MCI, and AD. Neurology 75:
143151.
35. Ewers M, Walsh C, Trojanowski JQ, Shaw LM, Petersen RC, et al. (2010)
Prediction of conversion from mild cognitive impairment to Alzheimers disease
dementia based upon biomarkers and neuropsychological test performance.
Neurobiol Aging.
36. Zhang D, Wang Y, Zhou L, Yuan H, Shen D (2011) Multimodal classification of
Alzheimers disease and mild cognitive impairment. Neuroimage 55: 856867.
45
Sensitivity and Specificity of Medial Temporal Lobe

Visual Ratings and Multivariate Regional MRI
Classification in Alzheimers Disease
Eric Westman1*., Lena Cavallin2,3., J-Sebastian Muehlboeck2, Yi Zhang2, Patrizia Mecocci4, Bruno
Vellas5, Magda Tsolaki6, Iwona Koszewska7, Hilkka Soininen8, Christian Spenger2, Simon Lovestone9,10,
Andrew Simmons9,10, Lars-Olof Wahlund1, for the AddNeuroMed consortium
1 Department of Neurobiology, Care Sciences and Society, Karolinska Institutet, Stockholm, Sweden, 2 Department of Clinical Science, Intervention and Technology,
Karolinska Institutet, Stockholm, Sweden, 3 Department of Radiology, Karolinska University Hospital, Stockholm, Sweden, 4 Institute of Gerontology and Geriatrics,
University of Perugia, Perugia, Italy, 5 INSERM U 558, University of Toulouse, Toulouse, France, 6 Aristotle University of Thessaloniki, Thessaloniki, Greece, 7 Medical
University of Lodz, Lodz, Poland, 8 University of Eastern Finland and University Hospital of Kuopio, Kuopio, Finland, 9 Kings College London, Institute of Psychiatry,
London, United Kingdom, 10 NIHR Biomedical Research Centre for Mental Health, London, United Kingdom
Abstract
Background: Visual assessment rating scales for medial temporal lobe (MTL) atrophy have been used by neuroradiologists
in clinical practice to aid the diagnosis of Alzheimers disease (AD). Recently multivariate classification methods for magnetic
resonance imaging (MRI) data have been suggested as alternative tools. If computerized methods are to be implemented in
clinical practice they need to be as good as, or better than experienced neuroradiologists and carefully validated. The aims
of this study were: (1) To compare the ability of MTL atrophy visual assessment rating scales, a multivariate MRI classification
method and manually measured hippocampal volumes to distinguish between subjects with AD and healthy elderly
controls (CTL). (2) To assess how well the three techniques perform when predicting future conversion from mild cognitive
impairment (MCI) to AD.
Methods: High resolution sagittal 3D T1w MP-RAGE datasets were acquired from 75 AD patients, 101 subjects with MCI and
81 CTL from the multi-centre AddNeuroMed study. An automated analysis method was used to generate regional volume
and regional cortical thickness measures, providing 57 variables for multivariate analysis (orthogonal partial least squares to
latent structures using seven-fold cross-validation). Manual hippocampal measurements were also determined for each
subject. Visual rating assessment of MTL atrophy was performed by an experienced neuroradiologist according to the
approach of Scheltens et al.
Results: We found prediction accuracies for distinguishing between AD and CTL of 83% for multivariate classification, 81%
for the visual rating assessments and 89% for manual measurements of total hippocampal volume. The three different
techniques showed similar accuracy in predicting conversion from MCI to AD at one year follow-up.
Conclusion: Visual rating assessment of the MTL gave similar prediction accuracy to multivariate classification and manual
hippocampal volumes. This suggests a potential future role for computerized methods as a complement to clinical
assessment of AD.
Citation: Westman E, Cavallin L, Muehlboeck J-S, Zhang Y, Mecocci P, et al. (2011) Sensitivity and Specificity of Medial Temporal Lobe Visual Ratings and
Multivariate Regional MRI Classification in Alzheimers Disease. PLoS ONE 6(7): e22506. doi:10.1371/journal.pone.0022506
Editor: Jerson Laks, Federal University of Rio de Janeiro, Brazil
Received April 14, 2011; Accepted June 22, 2011; Published July 21, 2011
Copyright: 2011 Westman et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
Funding: This study was supported by InnoMed (Innovative Medicines in Europe) an Integrated Project funded by the European Union of the Sixth Framework
program priority FP6-2004-LIFESCIHEALTH-5, Life Sciences, Genomics and Biotechnology for Health. AS and SL were supported by funds from the NIHR
Biomedical Research Centre for Mental Health at the South London (www.nihr.ac.uk), Maudsley NHS Foundation Trust and Institute of Psychiatry, Kings College,
London. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have the following competing interest: this study was supported by InnoMed (Innovative Medicines in Europe). There are no
patents, products in development or marketed products to declare. This does not alter the authors adherence to all the PLoS ONE policies on sharing data and
materials, as detailed online in the guide for authors.
Biomarkers of AD based on non-invasive in vivo methods are highly

desirable for diagnosis, monitoring disease progression and
evaluating disease-modifying treatment strategies. An ideal
biomarker would detect a fundamental feature of AD neuropathology, be diagnostically sensitive and specific, and produce
Introduction
Dementia is the third most common cause of death in society
today, exceeded only by cancer and cardiovascular disorders and
Alzheimers disease (AD) is the most common form of dementia.
46
July 2011 | Volume 6 | Issue 7 | e22506

MTL Visual Ratings and Multivariate Classification
study was approved by ethical review boards in each participating

country (local ethical review board at University of Perugia,
University of Toulouse, Aristotle University of Thessaloniki,
Medical University of Lodz, University of Eastern Finland and
University Hospital of Kuopio and Kings College London).
accurate and reproducible results [1]. Magnetic Resonance

Imaging (MRI), Positron Emission Tomography (PET), and
cerebrospinal fluid (CSF) measures all allow different aspects of
AD pathology to be studied. The new suggested research criterion
for AD is centered on a clinical core of early and significant
episodic memory impairment and at least one abnormal
biomarker from MRI, PET and CSF [2] and the new suggested
diagnostic criterion also utilize biomarkers [3].
Several groups including our own have proposed the use of
multivariate techniques for analyzing multiple regional measures
from MRI to aid diagnosis of AD and to predict future conversion
from the prodromal stages of the disease often referred to as mild
cognitive impairment (MCI) to AD. Previous studies have shown
that computerized methods give high prediction accuracies when
distinguishing between patient groups [4,5,6,7]. If computerized
methods based on MRI are to be useful in clinical practice then
they will need to be as good as or better than experienced
neuroradiologists. Kloppel et al. previously compared the
diagnostic accuracy of a computerized method (support vector
machines (SVM)) with radiological expertise, concluding that
SVM gives comparable results to a well-trained neuroradiologist
[8]. The study had the strength of using neuropathologically
confirmed images, however they did not use a validated and
widely used clinical rating scale and they studied relatively small
cohorts.
We recently used the multivariate method orthogonal partial
least squares to latent structures (OPLS) with multiple regional
volumes and cortical thickness measures as input to investigate
patterns of atrophy and prediction accuracy in two large
multicenter cohorts (AddNeuroMed and the Alzheimers Disease
Neuroimaging Initiative (ADNI)). The study included over a
thousand patients and we found prediction accuracies between
8387% when discriminating between AD and controls [9].
The aim of the current study is to compare the performance of
the OPLS multivariate technique with that of an experienced
neuroradiologist using data from the AddNeuroMed cohort (a part
of InnoMed, (Innovative Medicines in Europe), an Integrated
Project funded by the European Union Sixth Framework
programme) [10,11]. The method used for visual rating assessment
is the well established Scheltens method which uses a five point
scale to grade atrophy in the medial temporal lobe [12]. To our
knowledge this is the first comparison of the Scheltens visual rating
scale for assessment of AD with a computerized method.
Additionally our study uses a substantially larger cohort than the
earlier study by Kloppel et al, and we also include subjects with
mild cognitive impairment. For further evaluation we also aimed
to compare the visual rating assessment with manual hippocampal
measures. Manual measures have been used for AD diagnosis in
research for many years and this region is one part of the visual
rating assessment protocol described by Scheltens et al. However,
manual hippocampal measurements are time consuming and not
feasible in clinical practice. The different approaches were
compared in two steps, firstly by distinguishing between AD and
controls, and secondly by assessing how well the approaches
predicted conversion of MCI subjects at baseline to AD at one
year follow-up.
Study data and inclusion and diagnostic criteria

All patients originated from the AddNeuroMed project, part of
InnoMed (Innovative Medicines in Europe), a European Union
program designed to make drug discovery more efficient. The
project is designed to develop and validate novel surrogate
markers in Alzheimers disease (AD) and includes a human
neuroimaging strand [13,14] which combines MRI data with
other biomarkers and clinical data. Data was collected from six
different sites across Europe; University of Kuopio, Finland,
University of Perugia, Italy, Aristotle University of Thessaloniki,
Greece, Kings College London, United Kingdom, University of
Lodz, Poland and University of Toulouse, France. MRI images
from a total of 252 subjects were included in this study; 75 AD
patients, 101 MCI patients and 81 healthy controls. Demographics
of the cohort are given in Table 1. All AD and MCI subjects were
recruited from local memory clinics of the six participating sites
while the control subjects were recruited from non-related
members of the patients families, caregivers relatives or social
centres for the elderly. The inclusion and exclusion criteria were as
follows.
Alzheimers disease. Inclusion criteria: 1) ADRDA/NINCDS
and DSM- IV criteria for probable Alzheimers disease. 2) Mini
Mental State Examination score range between 12 and 28. 4) Age
65 years or above. Exclusion criteria: 1) Significant neurological or
psychiatric illness other than Alzheimers disease. This would
exclude patients with vascular dementia or large infarcts. 2)
Significant unstable systematic illness or organ failure. All AD
subjects had a Clinical Dementia Rating (CDR) scale score of 0.5
or above.
Mild Cognitive Impairment and Controls. Inclusion criteria:
1) Mini Mental State Examination score range between 24 and 30.
2) Geriatric Depression Scale score less than or equal to 5. 3) Age
65 years or above. 4) Medication stable. 5) Good general health.
Exclusion criteria: 1) Meet the DSM- IV criteria for Dementia. 2)
Significant neurological or psychiatric illness other than
Alzheimers disease. 3) Significant unstable systematic illness or
organ failure. The distinction between MCI and controls was
based on two criteria: 1) subject scores 0 on Clinical Dementia
Table 1. Subject characteristics.
AddNeuroMed

Ethics Statement
MCI
AD
Number
81
101
75
Female/Male
45/36
52/49
50/25
Age
73.666.3
74.065.8
74.266.0
Education
11.064.8
8.764.3
8.364.2
MMSE
29.061.2
27.261.6
21.364.6
CDR
0.5
1.160.5
ADAS1
3.461.5
5.361.2
6.661.5
Data are represented as mean 6 standard deviation. AD = Alzheimers disease,

MCI = Mild Cognitive Impairment, CTL = healthy control, Education in years,
MMSE = Mini Mental State Examination, ADAS1 = Word list non-learning (mean),
CDR = Clinical Dementia Rating.
Written consent was obtained where the research participant

had capacity, and in those cases where dementia compromised
capacity then assent from the patient and written consent from a
relative, according to local law and process, was obtained. This
CTL
47
Regional volume segmentation and cortical thickness

parcellation
Rating Scale = control. 2) Subject scores 0.5 on Clinical Dementia

Rating scale = MCI. For the MCI subjects it was preferable that
the subject and informant reported occurrence of memory
problems.
CDR, Mini-Mental State, and CERAD Cognitive Battery were
assessed for each subject. The CERAD Cognitive Battery was
replaced with the Alzheimers Disease Assessment Scale (ADAS
Cog) for the AD subjects. This cognitive test battery is specially
designed for AD trials [15]. Both the ADAS-cog and the CERAD
battery use the same 10-word recall task. The only difference is
that the scoring is inverted. The mean number of words not
recalled in the CERAD word list immediate recall task was
calculated. The variable obtained was named ADAS1, corresponding to the first subtest of ADAS-Cog. This was performed to
obtain comparable measures between groups.
We utilized a pipeline, developed by Fischl and Dale which

produces regional cortical thickness and volumetric measures
(Figure 1). Cortical reconstruction and volumetric segmentation
includes removal of non-brain tissue using a hybrid watershed/
surface deformation procedure [17], automated Talairach transformation, segmentation of the subcortical white matter and deep
grey matter volumetric structures (including hippocampus, amygdala, caudate, putamen, ventricles) [17,18,19] intensity normalization [20], tessellation of the grey matter white matter boundary,
automated topology correction [21,22], and surface deformation
following intensity gradients to optimally place the grey/white and
grey/cerebrospinal fluid borders at the location where the greatest
shift in intensity defines the transition to the other tissue class
[23,24,25]. Once the cortical models are complete, registration to a
spherical atlas takes place which utilizes individual cortical folding
patterns to match cortical geometry across subjects [26]. This is
followed by parcellation of the cerebral cortex into units based on
gyral and sulcal structure [27,28]. Left and right sided volumes and
thicknesses were averaged. The regional cortical thickness was
measured from 34 areas and the regional volumes were measured
from 23 areas. All volumetric measures from each subject were
normalized by the subjects intracranial volume. This segmentation
approach has been used for multivariate classification of Alzheimers disease and healthy controls [7], neuropsychological-image
analysis [29,30,31], imaging-genetic analysis [32,33] and biomarker
discovery [34].
MRI
Data acquisition for the AddNeuroMed study was designed to be
compatible with the Alzheimer Disease Neuroimaging Initiative
(ADNI) [16]. The imaging protocol for both studies included a high
resolution sagittal 3D T1-weighted MPRAGE volume (voxel size
1.161.161.2 mm3) and axial proton density/T2-weighted fast spin
echo images. The MPRAGE volume was acquired using a custom
pulse sequence specifically designed for the ADNI study to ensure
compatibility across scanners [16]. Full brain and skull coverage was
required for both of the latter datasets and detailed quality control
carried out on all MR images from both studies according to the
AddNeuroMed quality control procedure [13,14].
48
Figure 1. Representations of ROIs included as candidate input variables in the multivariate OPLS model. (A) Regional volumes. (B)
Regional cortical thickness measures.

Visual rating assessment for medial temporal lobe

atrophy
The 3D T1-weighted MPRAGE images were reoriented to an
oblique coronal orientation perpendicular to the AC-PC (anterior
commisure - posterior commisure) line suitable for both volumetric
and visual assessment. Visual assessment of the medial temporal
lobe atrophy was performed on a single MR-slice posterior to the
amygdala and the mamillary bodies positioned such that the
hippocampus, the pons and the cerebral peduncles are all covered
by the slice. The rating scheme used here was first proposed by
Scheltens et al. [12] and is based on a visual estimation of volume
of the medial temporal lobe. The visual assessment includes
hippocampus proper, dentate gyrus, subiculum, parahippocampal
gyrus, entorhinal cortex and surrounding CSF spaces such as the
temporal horns and choroid fissure. The right and left sides are
rated separately. Scores range from 0 (no atrophy) to 4 (end stage
atrophy) as detailed in Table 2 and Figure 2. For subjects ,75
years, a MTA score of 2 or more is considered abnormal, while for
subjects .75 years, a MTA score of 3 or more is considered
abnormal (http://www.radiologyassistant.nl/en/43dbf6d16f98d).
The rater (LC) was blinded to diagnosis, gender and age. Intrarater reliability of the visual assessment of the medial temporal
lobe atrophy was tested in 100 randomly selected subjects by
repeated assessment with an interval of one week. Intra-rater
reliability was 0.81 on right side and on left side 0.78. Weighted
kappa was 0.93 on both sides.
Manual segmentation of hippocampus

Manual measurements of hippocampal volume were performed
on a HERMES workstation (Nuclear Diagnostics, Stockholm,
Sweden). Each measurement was performed with constant
parameters by a neuroradiologist (YZ) who was blinded to clinical
information. A ROI tool was used within the HERMES Multimodality software package, to manually delineate the hippocampal
formation using previously defined anatomical landmarks [35].
Intra-rater reliability of the measurements was tested in 15
randomly selected subjects by repeated measurements with an
interval of one month. The intra class correlation coefficients
(ICC) of the measurements were .0.93. The total hippocampal
volume from each subject was normalized by the subjects
intracranial volume.
Figure 2. Visual assessment of the medial temporal lobe

atrophy was performed on a single MR-slice posterior to the
amygdala and the mamillary bodies. The was positioned so the
hippocampus, the pons and the cerebral peduncles were all visible. The
visual assessment included hippocampus proper, dentate gyrus,
subiculum, parahippocampal gyrus, entorhinal cortex and surrounding
CSF spaces such as temporal horn and choroid fissure. The right and left
side were rated separately. Scores range from 0 (no atrophy) to 4 (end
stage atrophy).
Multivariate data analysis

previously been used in several studies to analyze MR-data
[40,41,42,43,44]. OPLS and PLS give the same predictive
accuracy, but the advantage of OPLS is that the model created
to compare groups is rotated, which means that the information
related to class separation is found in the first component of the
model, the predictive component. The other orthogonal components in the model, if any, relate to variation in the data not
connected to class separation. Focusing the information related to
class separation on the first component makes data interpretation
easier [39].
Pre-processing was performed using mean centring and unit
variance scaling. Mean centring improves the interpretability of
the data, by subtracting the variable average from the data. By
doing so the data set is repositioned around the origin. Large
variance variables are more likely to be expressed in modeling
than low variance variables. Consequently, unit variance scaling
was selected to scale the data appropriately. This scaling method
calculates the standard deviation of each variable. The inverse
standard deviation is used as a scaling weight for each MRmeasure.
MRI measures were analyzed using orthogonal partial least

squares to latent structures (OPLS) [6,7,36,37,38,39], a supervised
multivariate data analysis method included in the software
package SIMCA (Umetrics AB, Umea, Sweden). A very similar
method, partial least square to latent structures (PLS) has
Table 2. Visual rating of the medial temporal lobe.
Scale
Width of
Choroid fissure
Width of
temporal horn
Hippocampal
thickness
qq
qqq
qq
QQ
qqq
qqq
QQQ
Scheltens et al., 1992.

49
at a fixed specificity (MCI-s predicted as CTL). We set the

specificity for all three methods for this comparison to that of the
visual assessment since this can not be changed and recalculated
the sensitivity and specificity for the other two methods.
The results from the OPLS analysis are visualized in a scatter

plot by plotting the predictive component, which contains the
information related to class separation. Components are vectors,
which are linear combinations of partial vectors and are
dominated by the input variables (x). The first and second
components are by definition orthogonal to each other and span
the projection plane of the points. Each point in the scatter plot
represents one individual subject. The predictive component
receives a Q2(Y) value that describes its statistical significance for
separating groups. Q2(Y) values .0.05 are regarded as statistically
significant [45], where
Q2 Y~1{PRESS=SSY
Results
Subject cohort
252 subjects were included in this study: 75 AD patients, 101
MCI patients and 81 control subjects. Using age as an x-variable
in the OPLS models did not have any effect on the predictive
power of the models separating the groups when all image
variables were included. Therefore, age was excluded from further
analysis. All MRI volumetric measures were normalised by
dividing by each subjects intracranial volume. As expected,
performance on the MMSE, CDR and ADAS1 was poorest
among AD patients and best among controls (Table 1). The MCI
group had scores between the AD and the control groups (Table 1).
where PRESS (predictive residual sum of squares) = S(yactual2

ypredicted)2 and SSY is the total variation of the Y matrix after
scaling and mean centring [45]. Q2(Y) is the fraction of the total
variation of the Ys (expected class values) that can be predicted by
a component according to cross validation (CV). Cross validation
is a statistical method for validating a predictive model which
involves building a number of parallel models. These models differ
from each other by leaving out a part of the data set each time.
The data omitted is then predicted by the respective model. In this
study we used seven fold cross-validation, which means that 1/7th
of the data is omitted for each cross-validation round. Data is
omitted once and only once. Variables were plotted according to
their importance for the separation of groups. The plot shows the
MRI measures and their corresponding jack-knifed confidence
intervals. Jack-knifing is used to estimate the bias and standard
error. Measures with confidence intervals that include zero have
low reliability [39]. Covariance is plotted on the y-axis, where
Covt,Xi ~tT Xi =N{1
50
OPLS modelling and quality

Two models were created using (1) total hippocampal volume
(2) automated regional volume and cortical thickness measures to
compare AD vs. controls. The first model using the total manual
hippocampal volume accounted for 100% of the variance of the
original data (R2(X)) and its cross validated predictability,
Q2(Y) = 0.61. The second model using regional MRI measures
resulted in one predictive component with R2(X) = 60% and cross
validated predictability Q2(Y) = 0.45.
Classification accuracy of the different techniques

The separation between patients with AD and controls and the
predictive power of the models Q2(Y) can be seen in Figure 3A
using automated regional volume and cortical thickness measures
as input. As can be observed there is a distinct separation between
AD and controls. This model resulted in a prediction accuracy of
82.7% (accuracy, sensitivity, specificity, positive likelihood ratio
and negative likelihood ratio given in Table 3). Figure 3B
illustrates the variables of importance for the distinction between
the two groups. The pattern of atrophy, including hippocampus,
amygdala and entorhinal cortex among other temporal lobe
regions, together with volume measures of CSF is as expected very
similar to previous analyses of the AddNeuroMed cohort using
regional MRI measures and an OPLS model [9]. Visual rating
assessment using the Scheltens scale resulted in a prediction
accuracy of 80.8%. Finally, total hippocampal volume yielded a
prediction accuracy of 89.1%. The best predictive result was
obtained from manual hippocampal measures closely followed by
the automated image pipeline with OPLS and lastly the visual
rating assessment.
where t is the transpose of the score vector t in the OPLS model, i

is the centered variable in the data matrix X and N is the number
of variables [39]. A measure with high covariance is more likely to
have an impact on group separation than a variable with low
covariance. MRI measures below zero in the scatter plot have
lower values in controls compared to AD subjects, while MRI
measures above zero are higher in controls compared to AD
subjects in the model.
Altogether 57 variables were used for OPLS analysis. No feature
selection was performed, meaning all measured variables were
included in the analysis. A model containing age was also created
to test if there were any significant differences between the
diagnostic groups in relation to the variable. We investigated
whether age would increase the predictive power of the models
using it as an x-variable.
Sensitivity and specificity were calculated from the crossvalidated prediction values of the OPLS models and for the visual
assessment. Finally, the positive and negative likelihood ratios
(LR+ = sensitivity/(1002specificity) and LR2 = (1002sensitivity)/
specificity) were calculated. A positive likelihood ratio between 5
10 or a negative likelihood ratio between 0.10.2 increases the
diagnostic value in a moderate way, while a value above 10 or
below 0.1 significantly increases the diagnostic value of the test
[46].
Finally the AD vs. CTL models were used as training sets to
investigate how well they could predict conversion from MCI to
AD after one year follow-up and how they compare to the visual
assessment. To easily compare the performance of the three
methods we also calculated the sensitivity (MCI-c predicted as AD)
Predicting conversion from MCI at baseline to AD at one

year follow-up
Finally, we wanted to investigate how the three different
approaches would predict conversion from MCI at baseline to AD
at one year clinical follow-up. All MCI subjects were classified as
either AD or control like using OPLS models (AD vs. CTL). The
visual assessment for the MCI subjects were performed in the same
way, using the same cut offs as described previously, resulting in an
assessment of abnormal or normal brain changes with respect to
age. The results are shown in Table 4 demonstrating that 68% of
the MCI converters (MCI-c) were classified as more AD-like and
68% of the MCI stable (MCI-s) classified as more control-like at
baseline using visual rating assessment. Using automated regional
MRI measures as input to the OPLS model, 74% of the MCI-c
5

Figure 3. OPLS cross validated score plots and MRI measures of importance for the separation between AD and CTL. (A) The scatter
plot visualises group separation and the predictability of the AD vs. CTL model. Each black square represents an AD subject and each gray circle a
control subject. Control subjects to the left of zero and AD subjects to the right of zero are falsely predicted. Q2(Y).0.05 (statistically significant
model). (B) Measures above zero have a larger value in controls compared to AD and measures below zero have a lower value in controls compared
to AD. A measure with a high covariance is more likely to have an impact on group separation than a measure with a low covariance. Measures with
jack knifed confidence intervals that include zero have low reliability.
subjects who converted to AD at one year follow-up were

predicted as more AD-like and 70% of the MCI-s were predicted
as more control-like at baseline. For the total hippocampal
volume, 79% of the MCI-c subjects who converted to AD at one
year follow-up were predicted as more AD-like and 54% of the
MCI-s were predicted as more control-like. When we set the
specificity (MCI-s predicted as CTL) to a fixed value (the
specificity of the visual assessment) to make the comparison of
the methods easier, the results were slightly altered (Table 4). The
best results were obtained from the OPLS model with automated
regional MRI measures as input, with 79% of MCI-c subjects who
converted to AD at one year follow-up predicted as more AD-like,
compared to 68% for the manual hippocampal volumes and

Scheltens visual assessment rating.
Discussion
Automated computerized MRI methods to aid in the diagnosis
of AD will only be implemented in clinical practice if they are
carefully investigated and validated. The aim of this study was to
further validate the OPLS technique with fully automated MRI
measures as input and compare it to the Scheltens scale for visual
assessment of medial temporal lobe atrophy and to that of total
hippocampal volume. To our knowledge this is the first time that
6
51
Table 3. Sensitivity/specificity and likelihood ratios for the different methods.
Sensitivity
Specificity
Accuracy
LR+
LR2
Visual assessment
78.7 (68.186.4)
82.7 (73.189.4)
80.8 (73.986.2)
4.6 (2.87.4)
0.26 (0.160.40)
Fischl and Dale
77.3(66.785.3)
87.7 (78.793.2)
82.7 (75.687.8)
6.3 (3.511.3)
0.26 (0.170.40)
Manual outlining
93.3 (85.397.1)
85.2(75.991.3)
89.1 (83.293.1)
6.3(3.710.7)
0.08 (0.030.18)
Confidence intervals within parentheses, LR+ = positive likelihood ratio and LR2 = negative likelihood ratio.
Classification accuracy of the different techniques
the Scheltens visual rating scale has been compared to a

computerized method. We wanted to investigate which approach
would distinguish between AD patients and controls with the
highest accuracy and best predict conversion from MCI to AD.
We have previously shown that combining a set of automated
measures of the brain together with manual measures of
hippocampus significantly improves the prediction accuracy using
OPLS [6]. Manual measures of different brain regions are time
consuming and operator dependent however and are hence not
regularly used in a clinical settings. Therefore we further
investigated the power of OPLS using only automated measures
as input (using the same volumes and cortical thickness measures
as used here) in two large cohorts (AddNeuroMed and Alzheimers
disease Neuroimaging Initiative (ADNI)). This was performed to
investigate if similar patterns of atrophy and prediction accuracy
could be obtained from two different large cohorts using the OPLS
model [9] and we found good comparability between the two
cohorts. We have also previously investigated the value of
combining magnetic resonance spectroscopy (MRS) with automated regional MRI measures using OPLS [7] which showed a
significant improvement compared with using either set of
measures individually. The next natural step was to compare the
OPLS technique with a well established visual rating assessment
scale such as the Scheltens scale, performed by an experienced
neuroradiologist, as we describe here.
The results suggest that the OPLS technique with fully

automated regional MRI measures as input performs better than
the visual rating assessments made by an experienced neuroradiologist. The sensitivity of the two methods is similar resulting in
identical negative likelihood ratios (77.3%, 0.26 and 78.7%, 0.26
respectively), while the specificity was higher for the OPLS
technique than the visual assessment (87.7% and 82.7%) yielding a
higher positive likelihood ratio (6.3 and 4.6). The overall accuracy
was higher for the OPLS technique compared to the visual rating
scale (82.7% and 80.8%). When the specificity was fixed for all
three methods at the value of the visual rating assessments, the
sensitivity of the OPLS analysis (79%) was better than the other
two methods (both 68%). Although the manually measured total
hippocampal volume still yielded the best prediction accuracy, the
time consuming nature of manual measures makes them
impractical in a clinical settings. Manual measurements can also
be operator dependent and it can be hard to compare such
measures across sites and within sites if different operators are
used. If only positive likelihood ratios are considered then the
OPLS method with fully automated image measures performs as
well as the manually measured total hippocampal volume. The
results from the OPLS method using the fully automated MRI
measures and the results from the hippocampal volumes are in line
with our earlier results from the AddNeuroMed cohort and have
been discussed and compared to that of other groups previously
[6,9]. Comparing the results from the current study with
previously published work we found only one similar study which
compared a computerized technique (SVM) with neuroradiologist
evaluations [8]. In this prior study, two small cohorts of AD
patients and controls were evaluated (40 subjects in the first cohort
and 28 subjects in the second cohort). The SVM and visual
assessments gave prediction accuracies of 95% vs. 88.8%
respectively, for the first cohort, and 92.9% vs. 82.5% for the
second cohort. The SVM results were better than the OPLS
results that we report here for both cohorts, with the result from
the visual rating assessment results higher for their first cohort than
our study, but similar to our study for their second cohort. The
cohorts included in the study by Kloppel et al. were neuropathologically confirmed and much smaller than the cohort investigated
in the present study, which is a likely explanation for the higher
prediction accuracy. Kloppel et al. also published another paper
[5] using pathologically confirmed data to distinguish between AD
and CTL using SVM. High prediction accuracies were again
obtained, up to 96% but the sample sizes were again smaller
(maximum 20 in each group). They also had a slightly larger
cohort (33 AD and 57 CTL) with probable mild AD subjects.
Using no feature selection, they obtained a prediction accuracy of
81.1% (sensitivity 60.6% and specificity 93.0%). Prediction
accuracies can sometimes be misleading, especially if one group
is much smaller than the other. In this present study using the
OPLS method with no feature selection on the automated
Table 4. MCI prediction.
52
Method
Number
AD-like
CTL-like
Visual assessment converters
19
68% (13)
32% (6)
Fischl and Dale converters
19
74% (14)
26% (5)
Manual hippocampal volume converters
19
79% (15)
21% (4)
Visual assessment non-converters
82
32% (26)
68% (56)
Fischl and Dale non-converters
82
30% (24)
70% (58)
Manual hippocampal volume nonconverters
82
46% (38)
54% (44)
Results for fixed specificity
Number
MCI-c
as AD
MCI-s
as CTL
Visual assessment
101
68%
68%
Fischl and Dale
101
79%
68%
Manual hippocampal volume
101
68%
68%
AD = Alzheimers disease, MCI = Mild Cognitive Impairment, CTL = healthy

control, MCI-c = MCI converters and MCI-s = MCI stable. To better compare the
performance of the three methods we also calculated the sensitivity (MCI-c
predicted as AD) at a fixed specificity (MCI-s predicted as CTL). We set the
specificity to that of the visual rating assessment and recalculated the sensitivity
and specificity of the other two methods used.

measure, we found an accuracy of 82.7% (sensitivity 77.3% and

specificity 87.7%). As can be observed the accuracy of the latter
study by Kloppel et al is very similar to ours, but their sensitivity is
much lower. It is important that computerized analysis techniques
are validated in large cohorts, since it is easy for such methods to
fix onto features that may be different between small cohorts, but
are not generalizable to larger cohorts. Although neuropathologically confirmed data is preferred when evaluating these types of
automated models, it is very difficult to obtain large neuropathologically confirmed datasets in practice. This is even more
pronounced when studying the prodromal stages of the disease.
MCI subjects are typically diagnosed approximately 1015 years
before death, which makes longitudinal follow-up very difficult
and the acquisition of large neuropathologically confirmed
datasets with recent MRI an exceedingly difficult endeavor. At
the moment it is necessary to choose between small data sets which
are neuropathologically confirmed, but potentially not representative of the heterogeneity and complexity of Alzheimers disease,
and larger datasets such as ours, which are more representative
but not neuropathologically confirmed. A further advantage of our
study, compared to that of Kloppel et al. is that the images
acquired in our study are ADNI compatible, making our findings
generalizable to other large ADNI compatible cohorts.
The new research and diagnostic criterion have a strong focus
on the use of biomarkers including MRI, PET and CSF [2,3].
Although individual markers such as MRI can be powerful in AD,
we believe that combining different markers is an attractive
approach, particularly if considering differential diagnosis amongst
different forms of dementia. Neurodegeneration in AD is
estimated to start 2030 years before the clinical diagnosis is
given [47] and so if diagnosis is to be made at the prodromal stage
of the disease then it may be necessary to combine several different
markers of disease. Previously we have shown that combining
regional MRI measures with magnetic resonance spectroscopy
results improves classification results, doubling the positive
likelihood ratio [7].
70%. Although the OPLS method gave the best results, subjects
diagnosed at baseline with MCI who are still classified as MCI at
one year follow up may subsequently convert to AD at a later stage
and thus assessing whether subjects will eventually convert to AD
may require longer follow up times. Further large studies with
longer follow up times are warranted to investigate this issue. To
make the comparison of the performance of the three methods
easier we calculated the sensitivity (MCI-c predicted as AD) at a
fixed specificity (MCI-s predicted as CTL). This yielded the best
sensitivity for the OPLS model with automated regional MRI
measures as input. This model predicted 79% of the MCI-c
subjects who converted to AD at one year follow-up as more ADlike, compared to 68% of the other two methods. Although the
number of converters is relatively small, a likely explanation for
these results is that combining multiple regions across the brain
aids in the prediction of MCI conversion.
Conclusion
Visual rating assessment of the medial temporal lobe gave
similar prediction accuracy to computerized multivariate classification and both accuracies are comparable to that of manual
hippocampal volume measurements. While manual hippocampal
volumes are a valuable research tool and have found utility in
clinical trials, they are not however practical for use in routine
clinical work. Our results demonstrate that computerized
multivariate classification is as good as expert radiological review
of MRI using a validated and widely used visual rating scale,
which suggests a potential future role for computerized methods as
a complement to clinical assessment. Improving classification
models by adding other biomarkers will hopefully increase the
predictive power of the models and improve our understanding of
the etiology of the disease. Two limitations of the current study are
however that the data is not neuropathologically confirmed and
that the follow-up of the MCI subjects was for one year.
In conclusion we believe that this study and previous work
[6,7,9] has shown that the OPLS model with automated MRI
measures as input has the potential to serve as a complement to
clinical assessment of AD, and to target appropriate populations
for clinical trials.
Predicting conversion from MCI at baseline to AD at one

year follow-up
At one year follow-up 19 of the 101 subjects with mild cognitive
impairment converted from MCI to AD while 82 remained stable.
One of the aims of this study was to compare how the three
methods performed in predicting conversion from MCI to AD.
The accuracy of predicting conversion varied from 6879%. Total
manual hippocampal volume gave the best results, followed by the
OPLS approach including automated MRI measures and finally
the visual rating assessment. It is however difficult to confidently
state that one method is better than the other due to the small
numbers of subject converting. The three methods predict 15/19,
14/19 and 13/19 converters respectively as more AD like, which
is a small absolute difference that has a larger impact on the
percentage accuracy. The accuracy of the different models in
correctly predicting stable MCI subjects varied between from 54
Acknowledgments
The authors thank the Gamla Tjanarinnor Foundation, the Swedish
Alzheimers Association, Swedish Brain Power, Health Research Council
of Academy of Finland and Stockholm Medical Image Laboratory and
Education (SMILE).
Conceived and designed the experiments: EW LC AS L-OW HS CS J-SM
YZ PM BV MT IK SL. Performed the experiments: AS HS PM BV MT
IK EW J-SM. Analyzed the data: EW LC YZ AS. Contributed reagents/
materials/analysis tools: EW LC AS L-OW HS CS J-SM YZ PM BV MT
IK SL. Wrote the paper: EW LC AS L-OW.
References
1. Kantarci K (2005) Magnetic resonance markers for early diagnosis and
progression of Alzheimers disease. Expert Rev Neurother 5: 663670.
2. Dubois B, Feldman HH, Jacova C, DeKosky ST, Barberger-Gateau P, et al.
(2007) Research criteria for the diagnosis of Alzheimers disease: revising the
NINCDS-ADRDA criteria. The Lancet Neurology 6: 734746.
3. McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR, Jr., et al.
(2011) The diagnosis of dementia due to Alzheimers disease: Recommendations
from the National Institute on Aging-Alzheimers Association workgroups on
diagnostic guidelines for Alzheimers disease. Alzheimers Dement 7: 263269.
4. Davatzikos C, Resnick SM, Wu X, Parmpi P, Clark CM (2008) Individual

patient diagnosis of AD and FTD via high-dimensional pattern classification of
MRI. Neuroimage 41: 12201227.
5. Kloppel S, Stonnington CM, Chu C, Draganski B, Scahill RI, et al. (2008)
Automatic classification of MR scans in Alzheimers disease. Brain 131:
681689.
Multivariate analysis of MRI data for Alzheimers disease, mild cognitive
impairment and healthy controls. Neuroimage 54: 11781187.
53
7. Westman E, Wahlund L-O, Foy C, Poppe M, Cooper A, et al. (2010)

Combining MRI and MRS to distinguish between Alzheimers disease and
healthy controls. J Alzheimers Dis 22: 171181.
8. Kloppel S, Stonnington CM, Barnes J, Chen F, Chu C, et al. (2008) Accuracy of
dementia diagnosis: a direct comparison between radiologists and a computerized method. Brain 131: 29692974.
9. Westman E, Simmons A, Muehlboeck S, Mecocci P, Vellas B, et al. (2011)
AddNeuroMed and ADNI: Similar Patterns of Alzheimers Atrophy and
Automated MRI Classification Accuracy in Europe and North America.
NeuroImage, In press.
10. Lovestone S, Francis P, Kloszewska I, Mecocci P, Simmons A, et al. (2009)
AddNeuroMed;The European Collaboration for the Discovery of Novel
Biomarkers for Alzheimers Disease. Annals of the New York Academy of
Sciences 1180: 3646.
11. Lovestone S, Francis P, Strandgaard K (2007) Biomarkers for disease
modification trialsthe innovative medicines initiative and AddNeuroMed.
J Nutr Health Aging 11: 359361.
12. Scheltens P, Leys D, Barkhof F, Huglo D, Weinstein HC, et al. (1992) Atrophy
of medial temporal lobes on MRI in probable Alzheimers disease and normal
ageing: diagnostic value and neuropsychological correlates. J Neurol Neurosurg
13. Simmons A, Westman E, Muehlboeck S, Mecocci P, Vellas B, et al. (2011) The
AddNeuroMed framework for multi-centre MRI assessment of longitudinal
changes in Alzheimers disease: experience from the first 24 months. Int J Geriatr
Psychiatry 26(1): 7582.
14. Simmons A, Westman E, Muehlboeck S, Mecocci P, Vellas B, et al. (2009) MRI
Measures of Alzheimers Disease and the AddNeuroMed Study. Annals of the
New York Academy of Sciences 1180: 4755.
15. Rosen WG, Mohs RC, Davis KL (1984) A new rating scale for Alzheimers
disease. Am J Psychiatry 141: 13561364.
The Alzheimers Disease Neuroimaging Initiative (ADNI): MRI methods.
J Magn Reson Imaging 27: 685691.
17. Segonne F, Dale AM, Busa E, Glessner M, Salat D, et al. (2004) A hybrid
approach to the skull stripping problem in MRI. Neuroimage 22: 10601075.
18. Fischl B, Salat DH, Busa E, Albert M, Dieterich M, et al. (2002) Whole brain
segmentation: automated labeling of neuroanatomical structures in the human
brain. Neuron 33: 341355.
19. Fischl B, Salat DH, van der Kouwe AJ, Makris N, Segonne F, et al. (2004)
Sequence-independent segmentation of magnetic resonance images. Neuroimage 23 Suppl 1: S6984.
automatic correction of intensity nonuniformity in MRI data. IEEE Trans Med
Imaging 17: 8797.
21. Fischl B, Liu A, Dale AM (2001) Automated manifold surgery: constructing
geometrically accurate and topologically correct models of the human cerebral
cortex. IEEE Trans Med Imaging 20: 7080.
22. Segonne F, Pacheco J, Fischl B (2007) Geometrically accurate topologycorrection of cortical surfaces using nonseparating loops. IEEE Trans Med
Imaging 26: 518529.
23. Dale AM, Fischl B, Sereno MI (1999) Cortical surface-based analysis. I.
Segmentation and surface reconstruction. Neuroimage 9: 179194.
24. Dale AM, Sereno MI (1993) Improved Localizadon of Cortical Activity by
Combining EEG and MEG with MRI Cortical Surface Reconstruction: A
Linear Approach. Journal of Cognitive Neuroscience 5: 162176.
25. Fischl B, Dale AM (2000) Measuring the thickness of the human cerebral cortex
from magnetic resonance images. Proc Natl Acad Sci U S A 97: 1105011055.
26. Fischl B, Sereno MI, Tootell RB, Dale AM (1999) High-resolution intersubject
averaging and a coordinate system for the cortical surface. Hum Brain Mapp 8:
272284.
27. Desikan RS, Segonne F, Fischl B, Quinn BT, Dickerson BC, et al. (2006) An
automated labeling system for subdividing the human cerebral cortex on MRI
scans into gyral based regions of interest. NeuroImage 31: 968980.
28. Fischl B, van der Kouwe A, Destrieux C, Halgren E, Segonne F, et al. (2004)
Automatically parcellating the human cerebral cortex. Cereb Cortex 14: 1122.
29. Liu Y, Paajanen T, Zhang Y, Westman E, Wahlund L-O, et al. (2011)
in differentiating AD, MCI and control groups-The AddNeuroMed study.
30. Liu Y, Paajanen T, Zhang Y, Westman E, Wahlund L-O, et al. (2010) Analysis
of regional MRI volumes and thicknesses as predictors of conversion from mild
13751385.
31. Tunnard C, Whitehead D, Hurt C, Wahlund L-O, Mecocci P, et al. (2011)
Apathy and cortical atrophy in Alzheimers disease. Int J Geriatr Psychiatry 26:
741748.
32. Liu Y, Paajanen T, Westman E, Wahlund L-O, Simmons A, et al. (2010) Effect
of APOE epsilon4 allele on cortical thicknesses and volumes: the AddNeuroMed
study. J Alzheimers Dis 21: 947966.
33. Liu Y, Paajanen T, Westman E, Zhang Y, Wahlund L-O, et al. (2010) APOE
epsilon2 allele is associated with larger regional cortical thicknesses and volumes.
Dement Geriatr Cogn Disord 30: 229237.
34. Thambisetty M, Simmons A, Velayudhan L, Hye A, Campbell J, et al. (2010)
Association of plasma clusterin concentration with severity, pathology, and
progression in Alzheimer disease. Arch Gen Psychiatry 67: 739748.
35. Pantel J, OLeary DS, Cretsinger K, Bockholt HJ, Keefe H, et al. (2000) A new
method for the in vivo volumetric measurement of the human hippocampus with
high neuroanatomical accuracy. Hippocampus 10: 752758.
36. Bylesjo M, Eriksson D, Kusano M, Moritz T, Trygg J (2007) Data integration in
plant biology: the O2PLS method for combined modeling of transcript and
metabolite data. Plant J 52: 11811191.
37. Johan Trygg SW (2002) Orthogonal projections to latent structures (O-PLS).
Journal of Chemometrics 16: 119128.
38. Rantalainen M, Cloarec O, Beckonert O, Wilson ID, Jackson D, et al. (2006)
Statistically integrated metabonomic-proteomic studies on a human prostate
cancer xenograft model in mice. J Proteome Res 5: 26422655.
39. Wiklund S, Johansson E, Sjostrom L, Mellerowicz EJ, Edlund U, et al. (2008)
Visualization of GC/TOF-MS-based metabolomics data for identification of
biochemically interesting compounds using OPLS class models. Anal Chem 80:
115122.
40. Levine B, Kovacevic N, Nica EI, Cheung G, Gao F, et al. (2008) The Toronto
traumatic brain injury study: Injury severity and quantified MRI. Neurology 70:
771778.
41. McIntosh AR, Lobaugh NJ (2004) Partial least squares analysis of neuroimaging
data: applications and advances. Neuroimage 23: S250S263.
42. Oberg J, Spenger C, Wang FH, Andersson A, Westman E, et al. (2007) Age
related changes in brain metabolites observed by (1)H MRS in APP/PS1 mice.
Neurobiol Aging 29(9): 142333.
43. Westman E, Spenger C, Oberg J, Reyer H, Pahnke J, et al. (2009) In vivo 1Hmagnetic resonance spectroscopy can detect metabolic changes in APP/PS1
mice after donepezil treatment. BMC Neurosci 10: 33.
44. Westman E, Spenger C, Wahlund L-O, Lavebratt C (2007) Carbamazepine
treatment recovered low N-acetylaspartate+N-acetylaspartylglutamate (tNAA)
levels in the megencephaly mouse BALB/cByJ-Kv1.1mceph/mceph. Neurobiology of Disease 26: 221228.
45. Eriksson L, Johansson E, Kettaneh-Wold N, Trygg J, Wiksrom C, et al. (2006)
Multi- and Megavariate Data Analysis (Part I -Basics and Principals and
Applications). Umea: Umetrics AB.
46. Qizilbash N, Schneider L, Chui H, Tariot P, Brodaty H, et al. (2002)
Evindenced-based Dementia Pratice. Oxford, UK: Blackwell Publishing. pp
2023.
47. Blennow K, de Leon MJ, Zetterberg H (2006) Alzheimers disease. The Lancet
368: 387403.
54
MRI Markers for Mild Cognitive Impairment:

Comparisons between White Matter Integrity and Gray
Matter Volume Measurements
Yu Zhang1,2*, Norbert Schuff1,2, Monica Camacho1,2, Linda L. Chao1,2, Thomas P. Fletcher3,
Kristine Yaffe4, Susan C. Woolley5, Catherine Madison5, Howard J. Rosen6, Bruce L. Miller6,
Michael W. Weiner1,2
1 Center for Imaging of Neurodegenerative Diseases, San Francisco VA Medical Center, San Francisco, California, United States of America, 2 Department of Radiology and
Biomedical Imaging, University of California San Francisco, San Francisco, California, United States of America, 3 The Scientific Computing and Imaging Institute at the
University of Utah, Salt Lake City, Utah, United States of America, 4 Department of Psychiatry, Neurology and Epidemiology, San Francisco VA Medical Center and
University of California San Francisco, San Francisco, California, United States of America, 5 Department of Neurology, California Pacific Medical Center, San Francisco,
California, United States of America, 6 Department of Neurology, University of California San Francisco, San Francisco, California, United States of America
Abstract
The aim of the study was to evaluate the value of assessing white matter integrity using diffusion tensor imaging (DTI) for
classification of mild cognitive impairment (MCI) and prediction of cognitive impairments in comparison to brain atrophy
measurements using structural MRI. Fifty-one patients with MCI and 66 cognitive normal controls (CN) underwent DTI and
T1-weighted structural MRI. DTI measures included fractional anisotropy (FA) and radial diffusivity (DR) from 20
predetermined regions-of-interest (ROIs) in the commissural, limbic and association tracts, which are thought to be involved
in Alzheimers disease; measures of regional gray matter (GM) volume included 21 ROIs in medial temporal lobe, parietal
cortex, and subcortical regions. Significant group differences between MCI and CN were detected by each MRI modality: In
particular, reduced FA was found in splenium, left isthmus cingulum and fornix; increased DR was found in splenium, left
isthmus cingulum and bilateral uncinate fasciculi; reduced GM volume was found in bilateral hippocampi, left entorhinal
cortex, right amygdala and bilateral thalamus; and thinner cortex was found in the left entorhinal cortex. Group
classifications based on FA or DR was significant and better than classifications based on GM volume. Using either DR or FA
together with GM volume improved classification accuracy. Furthermore, all three measures, FA, DR and GM volume were
similarly accurate in predicting cognitive performance in MCI patients. Taken together, the results imply that DTI measures
are as accurate as measures of GM volume in detecting brain alterations that are associated with cognitive impairment.
Furthermore, a combination of DTI and structural MRI measurements improves classification accuracy.
Citation: Zhang Y, Schuff N, Camacho M, Chao LL, Fletcher TP, et al. (2013) MRI Markers for Mild Cognitive Impairment: Comparisons between White Matter
Integrity and Gray Matter Volume Measurements. PLoS ONE 8(6): e66367. doi:10.1371/journal.pone.0066367
Editor: Yong Fan, Institution of Automation, CAS, China
Received October 10, 2012; Accepted May 7, 2013; Published June 6, 2013
Copyright: 2013 Zhang et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
Funding: This research was supported in part by National Institutes of Health/National Institute on Aging grants P01AG19724, P01AG12435 and grant from NIH/
National Center for Research Resources P41RR023953, which were administered by the Northern California Institute for Research and Education, and with
resources of the Veterans Affairs Medical Center, San Francisco, California. The funders had no role in study design, data collection and analysis, decision to
publish, or preparation of the manuscript.
Introduction
attracted considerable interest, in part because of the low risk, noninvasive examinations and broad availability nowadays.
Aside from MRI measurements of brain atrophy [614],
diffusion tensor imaging (DTI), a variant of MRI that is sensitive
to microstructural tissue changes, has also shown promise in
characterizing MCI [1524]. DTI studies in MCI generally rely
on diffusion summary measures, such as fractional anisotropy (FA),
an index of the spatial directionality of tissue water diffusion and
mean diffusivity (MD), a measure of the diffusion magnitude.
These measures uniquely detect abnormalities in white matter
(WM) regions. In particular, FA alterations in MCI though often
subtle [15,17,2528] have been found in limbic circuits
(including posterior and parahippocampal tracts and fornix)
[19,2932] and posterior callosal areas [33,34]. To gain further
insight into WM alterations in MCI, DTI studies have computed
Mild cognitive impairment (MCI) is a transitional stage between

normal aging and early Alzheimers disease (AD), though not all
individuals with a diagnosis of MCI convert to AD [1]. Accurate
identification of patients with MCI who are at increased risk of
developing AD may facilitate timely initiation of treatments and
possible prevention of dementia. While clinical assessments remain
the standard diagnostic criteria for MCI, various imaging and
biomarker measures have been proposed as complementary tests
to improve diagnosis. Proposed measures include: MRI measured
atrophy of the hippocampus and entorhinal cortex [2], 18Ffluorodeoxyglucose PET measured temporoparietal cortical hypometabolism, brain amyloid PET [3], and abnormal level of
cerebrospinal fluid markers (amyloid-b42, total tau, and phosphorylated tau) [4,5]. Among these proposed markers, MRI has
June 2013 | Volume 8 | Issue 6 | e66367
55
Structural and Diffusion MRI in MCI
subjects had predominant memory impairments) and nonamnestic MCI (naMCI, if the MCI subjects had predominant
non-memory impairments, such as executive, language, visual,
behavioral, motor, or other deficits). Sixty-six cognitively normal
subjects, whose neuropsychological scores were no worse than one
standard deviation below the mean on standardized tests, but also
not better than 2 standard deviations above the mean were
included as controls. Exclusion criteria for all participants included
any poorly controlled illness, use of medication or recreational
drugs that could affect brain function, a history of brain trauma,
brain surgery, ischemic events, or skull defects.
axial (DA) and radial (DR) diffusivity data in addition to FA. DR

reflects the diffusion magnitude along the orthogonal directions of
the diffusion tensor. In particular, an increase in DR has been
associated with demyelination and axonal degeneration [35].
Some DTI studies using DR have demonstrated distinct zones of
WM alterations in MCI where changes were barely seen with FA
[22,36]. However, aside from a previous work [19] in our
laboratory, few studies [25,37,38] have compared the accuracy of
DTI measurements with structural imaging measurements in
characterizing brain alterations in MCI. It is not clear whether the
various indices of DTI are superior to, or provide complementary
information to structural MRI measures of brain atrophy for
accurately classifying MCI and predicting cognitive deficits.
The aims of the study were: 1) to determine to what extent GM
atrophy measured by structural MRI and WM alterations
measured by DTI in prominent brain regions affected in AD are
also associated with MCI; 2) to test whether WM alterations based
on DTI improve the classification of MCI, in comparison to
classifications based on GM volume; 3) to compare the accuracies
of using GM volume and DTI measures for prediction of cognitive
performance in MCI.
MRI Acquisitions
All scans were preformed on a 4 Tesla (Bruker/Siemens) MRI
system with a single housing birdcage transmit and 8-channel
receiver head coil. T1-weighted images were obtained using a 3D
volumetric magnetization prepared rapid gradient echo
(MPRAGE) sequence with TR/TE/TI = 2300/3/950 ms, 7degree flip angle, 1.061.061.0 mm3 resolution, 157 continuous
sagittal slices. In addition, FLAIR (fluid attenuated inversion
recovery) images (TR/TE/TI = 5000/355/1900 ms) and T2weighted images (TR/TE = 4000/30 ms) were acquired for
clinical reading. DTI was acquired based on a dual-refocused
spin-echo EPI sequence supplemented with twofold parallel
imaging acceleration (GRAPPA) [43] to reduce susceptibility
distortions. Other DTI parameters were TR/TE = 6000/77 ms,
field of view 2566224 cm, 1286112 matrix size, yielding
262 mm2 in-plane resolution, and 40 contiguous slices each
3 mm thick. One reference image (b = 0) and six diffusionweighted images (b = 800 s/mm2, along 6 noncollinear directions)
were acquired.

Subjects and Clinical Assessments
56
One hundred and seventeen subjects participated in the study.

Subjects were recruited between May 2005 and October 2010
either by memory clinics in the San Francisco Bay Area, including
the Memory Disorders Clinic at the San Francisco Veterans
Affairs Medical Center, the Memory and Aging Center at the
University of California, San Francisco, and the Memory Clinic at
the California Pacific Medical Center or by posting flyers and
advertisements in local newspapers. All subjects gave written
informed consent, approved by the institutional review boards of
the University of California San Francisco and the San Francisco
VA Medical Center.
All subjects received the same research MRI examinations,
medical and neurological examinations and neuropsychological
testing. Domains assessed during baseline neuropsychological
testing included: general cognitive ability (i.e., Mini-Mental State
Examination, (MMSE) [39]), global cognitive functioning (i.e.,
Clinical Dementia Rating, (CDR) [40]), phonemic verbal fluency
(D words) and semantic fluency (Animal Naming) [41] California
Verbal Learning Test - Short Form (CVLT II) [42] including
Immediate Recall (sum of trials 14 correct scores with total 36
items), 9-item Long Delay Free Recall and Long Delay Cued
Recall. To facilitate predictions of accuracies, the 3 CVLT II
recall scores (immediate recall, long free and cued recalls) were
further composited as one CVLT score because they are
correlated with each other.
None of the subject met the clinical diagnostic criteria for
dementia. Fifty-four subjects were diagnosed with MCI by
clinicians based on clinical criteria established by Peterson et al.,
or by the Alzheimers Disease Cooperative Study (ADCS), or if the
subject met the MCI core diagnostic features, which include:
subjective memory complaints and memory difficulties that are
qualified by an informant; memory impairment which exceeds
normal aging, as indexed by low cognitive performance on one or
more neuropsychological tests which assess learning or recall (for
example, delayed recall, word list); a MMSE score greater than or
equal to 24/30; a CDR score of 0.5 (but not demented); intact
activities of daily living; and no dementia. All MCI subjects, with
the exception of three whose diagnostic details were missing, were
further divided into two groups: amnestic MCI (aMCI, if the MCI
GM Volume Data Processing

Automated cortical volume measures, cortical parcellation, and
subcortical segmentation were performed with FreeSurfer software
package, version 4.5 (surfer.nmr.mgh. harvard.edu/fswiki).
For a full description of the FreeSurfer processing steps, see
Fischl et al [44,45]. In the first step, each T1-weighted volume was
first corrected for head motion, then transformed into Talairach
space using affine transformations (12 degrees of freedom),
corrected for intensity non-uniformity, and separated from nonbrain tissue. The remaining brain image volume was further
normalized to match the intensity of a probabilistic atlas with 40
anatomical labeled regions of interest (ROI) per hemisphere before
the atlas image was nonlinearly warped to the individual brain
space to facilitate atlas-based tissue segmentation and anatomic
labeling of subcortical structures, brain stem, cerebellum, and
cerebral cortex. In the second step, FreeSurfer was used to
generate topological cortical surface representations per hemisphere with theoretical surface bounded by the GM interface to
white matter at the inner side and CSF at the outer side. Each
hemispheres cortical surface representation was mapped to the
spherical coordinate system of the anatomical-labeled atlas,
allowing an automated anatomical parcellation of cortex into
gyral regions. The surface parcellation was then extended to GM
volume, yielding parcellation of GM tissue sheet, volume and
averaged thickness in each cortical and subcortical ROI. The
anatomical accuracy of FreeSurfer processing was visually
reviewed by trained readers with neuroanatomic knowledge but
no corrections were performed by manual editing to avoid rater
bias. Instead, data that did not pass the quality control of
FreeSurfer processing were not included in the study.
Based on previous MRI reports of prominent brain alterations
in AD and MCI [4648], GM volumes were evaluated in 21
2

anatomical ROIs, including 3 callosal regions (anterior, middle

and posterior corpus callosum), 8 medial temporal regions
(bilateral hippocampus, entorhinal cortex, parahippocampal cortex and amygdala), 6 parietal regions (bilateral precuneus,
posterior and isthmus cingulate gyrus) and 4 subcortical regions
(bilateral thalamus and putamen) (Figure 1A). Although several
other ROIs, such as inferior, middle and superior temporal cortex
as well as inferior parietal cortex have been reported in the
literature, we were unable to obtain reliable values from these
regions across all subjects with the automated Freesurfer
procedure due to limited image quality. Thus the study was
limited in the above 21 ROIs in which FreeSurfer measurements
were available for all participants. Regional values of the GM
volume were normalized to total intracranial volume (ICV), to
account for variations in head size.
all individual FA images and the JHU ICBM-DTI-81 FA and

labeled atlas were spatially normalized onto a common space of
the averaged FA template. 4) After that, a reversed warping
procedure was applied to assign the JHU ICBM-DTI-81 labeled
atlas from the DARTEL common space to each individual FA
space. 5) In subjects individual space, FA and DR were extracted
from all labeled ROIs. To avoid inclusion of surrounding GM or
CSF tissues, DTI values were obtained only in voxels with FA
values greater than 0.20. Furthermore, data quality was reviewed
and ROIs which contained visual mis-registration or WM
hyperintensities were excluded. The decision to exclude ROIs
with dominant WM hyperintensities was based on detection of a
skewed FA distribution within the ROI toward very low FA values
(,0.2) in combination with the visual inspection of the
corresponding structural MRI for WM lesions.
Based on previous DTI reports [32,50], FA and DR of 20
anatomical ROIs were measured, including 3 regions of the
commissural tracts (genu, body and splenium of the corpus
callosum), 9 regions of the limbic tracts (fornix and bilateral Fornix
(cres) with Stria terminalis, posterior and isthmus cingulum, and
hippocampal cingulum) and 8 regions of the association tracts
(bilateral inferior and superior longitudinal fasciculi, inferior
fronto-occipital fasciculi and bilateral uncinate fasciculi)
(Figure 1B).
DTI Data Processing

Preprocessing of the DTI data included skull-striping, motion
and eddy-current correction with FSL package (http://www.
fmrib.ox.ac.uk/fsl/), and geometric distortion correction based on
a variational matching algorithm [49]. After FA and DR images
were created for each participant, these images underwent an
automated region of interest (ROI) extraction procedure: 1) All FA
images were resliced to the SPM8 white matter (WM) template
using a rigid-body transformation (dimensions: 12161456121
voxels, resolution: 1.5 mm3). 2) The JHU ICBM-DTI-81 atlas
package (cmrm.med.jhmi.edu/), which includes a probabilistic FA
atlas and anatomical labels of 44 deep WM ROIs from the DTI
data of 81 subjects, was imported into SPM8, and also resliced to
the SPM8 WM template using a rigid-body transformation. 3) The
resliced FA images were warped nonlinearly to an intensityaveraged FA template generated from all subjects using a
diffeomorphic registration algorithm (DARTEL). Consequently,
Statistics
Differences between patients and controls were tested using a ttest for each ROI and MRI modality after effects of age and
gender were linearly regressed out. For the rest of the analysis, FA,
DR, and GM volume measures were centered and normalized to
their respective standard deviations (i.e. transformed to z-scores) to
eliminate scaling differences across the ROIs and MRI modalities.
In addition, a non-parametric test using permutations of the
Figure 1. GM and WM parcellations. A. Automated parcellation of 21 cortical and subcortical ROIs for GM measurement, performed by Freesurfer
software. B. Automated parcellation of 20 deep WM ROIs for DTI measurement, performed by SPM8.
57
Table 1. Subjects and clinical characteristics.
Control
MCI subtypes
MCI
naMCI
Subject Number
66
51
Age (years)
67.2610.0
72.868.7
Sex
31 M : 35 F
Years of Education*
16.762.4
MMSE*
29.461.0
27.861.9
Immediate Recall (Trials 14)
28.264.1
19.166.8
Long Delay Free Recall
7.1261.5
4.5262.7
Long Delay Cued Recall
7.4561.3
4.6562.4
Verbal Fluency* (D-words)
14.665.6
13.466.0
Semantic Fluency* (Animal Naming)
21.666.1
15.766.2
aMCI
17
31
71.9610.9
73.167.5
30 M : 21 F
9M:8F
20 M : 11 F
18.6612.4
22.1621.5
17.562.4
28.561.7
27.561.9
21.666.0
17.166.8
5.7661.9
3.9362.8
5.9461.9
3.8762.5
`
`
`
`
ICV (cm )
10766125
10916147
Total GM/ICV
0.4060.04
0.3760.04
Total WM/ICV
0.4160.05
0.3960.05
14.365.9
12.966.1
18.967.0
13.965.3
10466111
11086164
0.3860.03
0.3660.04
0.3960.05
0.3960.05
*6 subjects years of education was missing. 4 subjects MMSE was missing. 28 subjects verbal fluency and semantic fluency were missing. Note, smaller scores of
neurocognitive measures indicate greater impairment.
Significance of group differences between paired groups (MCI vs. Control, aMCI vs. naMCI):
{
0.05,p#0.001,
`
p,0.001.
less total GM volume (normalized to ICV) than CN, while

differences in total WM volume (also normalized to ICV) were not
significant.
diagnostic labels was applied to determine whether the heterogeneity of MCI has to be taken into account for findings of
differences between uniform MCI and CN. The concept of false
discovery rate (FDR) was used to account for the multiple
comparisons problem [51]. An adjusted level of p,0.05 was
selected as threshold of significance. The classification accuracy of
FA, DR, and GM volume was tested using regularized logistic
regressions with all respective ROIs included simultaneously as
factors. An L2-norm regularization term (Ridge regression) was
used to control for overfitting [52]. The classification accuracy of
FA, DR, and GM were then compared pairwise by testing
differences in the area under the curve of an operator characteristics analysis (ROC) based on bootstraping (using the roc.test
function in the pROC package of R, URL: http://www.r-project.
org/index.html). Similarly, to test the accuracy of FA, DR, and
GM volume in predicting neurocognitive performance, regularized linear regressions were used with cognitive scores as outcomes
and with all respective ROIs included simultaneously as factors.
Accuracy was expressed as the root mean square error (RMSE)
between predicted and clinical values, where a smaller RMSE
indicating a more accurate prediction. Differences in RMSE
distributions of FA, DR, and GM volume where compared using ttests. Lastly, classifications and predictions were performed with 5fold cross-validation and bootstrapping to estimate reliability in
terms of 95% confidence intervals (CI).
Group differences based on GM volume and DTI

measurements
Mean and standard error of group differences in MRI
measurements for all ROIs are depicted in Figure 2A, separately
for diffusion and volume. As for DTI, mean FA was significantly
reduced in MCI as compared to controls in the splenium of the
corpus callosum, the left isthmus cingulum and the fornix.
Interestingly, FA of the left inferior longitudinal fasciculus was
increased in MCI. DR was increased in MCI as compared to
controls in the splenium of the corpus callosum, the left isthmus
cingulum and bilaterally in the uncinate fasciculi, consistent with
reduced FA in these regions. As for GM volume, the most
prominent volume losses in MCI patients compared to controls
involved bilateral hippocampi, the left entorhinal cortex, the right
amygdala, and the thalamus bilaterally. The most prominent
regions of cortical thinning in MCI included the left entorhinal
cortex. The effect sizes of significant group differences based on
FA and DR were similar (effect size = 0.420.61) to those based
on GM volumes and thickness (effect size = 0.420.61). Differences between the left and the right side of the brain were not
significant (p.0.65).
Differences between MCI subgroups in MRI measurements for
all ROIs are depicted in Figure 2B, separately for diffusion and
volume. As for DTI, significant differences between the MCI
subgroups were found in bilateral posterior cingulum, right
superior longitudinal fasciculus and left inferior longitudinal
fasciculus. In these regions, DTI (FA/DR) values of the naMCI
group were more abnormal than those of the aMCI group, while
the values of the aMCI group was on average not different from
those of controls. The difference between the two MCI subgroups
was dominated by better than normal values in the aMCI group.
Results
Demographic and Clinical Characteristics
58
The clinical and demographic characteristics of the subjects as

well as global brain volumes are summarized by diagnostic group
in Table 1. Differences in gender and years of education between
MCI patients and CN subjects were not significant. However,
MCI patients were older than CN subjects and performed worse
on most neuropsychological tests as expected, with the exception
of verbal fluency (D-words). The MCI patients had, on average,

59
Figure 2. Mean differences of the DTI and GM measures between MCI and control. A. Mean differences and standard errors of regional DTI
and GM measures, expressed as Z-scores, between MCI patients and controls for each ROI. Abbreviations:CC = corpus callosum; Post. CG = posterior
cingulum; Isth. CG = isthmus cingulum; FX-ST = fornix (cres) and stria terminalis; IFO = inferior fronto-occipital fasciculus; SLF = superior
longitudinal fasciculus; ILF = inferior longitudinal fasciculus. B. Mean differences and standard errors of regional DTI and MRI measures between
aMCI, naMCI group and controls for each ROI. DTI and GM measures in ROIs with significant heterogeneities between aMCI and naMCI groups were
labeled as #.
Predictions of Cognitive Deficits
As for GM measures, significant differences between the MCI

subgroups were found in the right hippocampus, left entorhinal
cortex, bilateral amygdale, and the right parahippocampal cortex.
In these regions, the difference between the two MCI subgroups
was dominated by more GM loss and cortical thinning in the
aMCI group. Hence, variations in the aMCI group dominated
differences between MCI overall and controls.
The accuracy of each FA, DR, or GM volume in predicting

clinical scores of cognitive deficits is illustrated in Figure 5,
showing the distribution of root mean square errors for each
measure. The three imaging measures achieved similar levels of
accuracy for predicting MMSE and CVLT.
Discussion
Relations between DTI and GM volume in MCI
The main findings of this study are: 1) MCI is associated with

significant alterations of FA and DR as well as GM volume loss in
specific regions, in agreement with previous findings. 2) FA and
DR achieve better classification accuracy than GM volume alone,
and using DTI and GM volume together further improves
classification. 3) FA, DR and GM volume are similarly accurate in
predicting cognitive performance. Taken together, the results
imply that DTI of WM alterations is as accurate as structural MRI
of GM volume in detecting brain alterations that associated with
cognitive impairment. Furthermore, a combination of DTI and
structural MRI measures improves accuracy.
DTI has been used in more than a hundred MRI reports of
MCI and AD studies. However, the findings have been
inconsistent [53], unlike structural MRI reports of medial
temporal lobe atrophy in MCI and AD. Different methodologies
in selecting regions of interest are a potential source of
inconsistency in DTI findings. We used an automated region
selection method to eliminate rater dependent bias and also
evaluated all selected regions simultaneously for analysis to gain
statistical power. Our findings of reduced FA and increased DR in
Relationships between regional DTI alterations in white matter

and atrophy of GM in MCI were investigated using Pearson
correlation coefficients. The significance of regional correlations
are illustrate in Figure 3, separate for FA and DR measures in
relation to GM volumes. Significant DTI-volume correlations
were found between the limbic structures and the corpus callosum,
whereas other regions showed no significant relationships.
Group classifications
Classification accuracy of MCI and control subjects based on
either FA, DR, GM volume alone or in combinations is
summarized in Table 2. The corresponding receiver operating
characteristic curves of each classification are illustrated in
Figure 4. The table indicates that classifications based on FA or
DR alone were significant (lower bound of the confidence interval
.50%) in contrast to classifications based on GM volumes that
were not better than chance. However, using either DR or FA
together with GM volume further improved accuracy.
60
Figure 3. The correlations between DTI and GM volumes in MCI patients. The p value maps of the Pearsons correlation between DTI values
and GM volumes. The green and warmer colors indicate significant correlations.

in regions of the splenium, fornix, isthmus cingulum and uncinate

fasciculus in MCI highlight a network of brain regions which are
known to be vulnerable to AD pathology. The pattern of these
regional DTI alterations suggests that regional degradation of WM
matter integrity could be a sign of incipient AD pathology in MCI.
An interesting finding is increased FA and decreased DR in the
inferior longitudinal fasciculus in MCI. An increase in FA has been
reported in other pathological conditions [59]. It has been
proposed that reduced dendritic branching of the posterior
parietal region may lead to an FA increase, though this hypothesis
needs to be tested. However, increased FA could also be
methodological artifacts induced by variations in the macroscopic
architecture of the long association fibers [60]. The finding of
increased FA of the inferior longitudinal fasciculus in MCI
requires further investigations. In particular, DTI studies with
more than six diffusion encoding directions need to be performed
to achieve higher accuracy in measuring FA. In addition, the
finding needs to be replicated in a separate population of MCI.
In contrast to DTI findings, our findings of GM atrophy in the
medial temporal lobe and subcortical regions are generally in line
with numerous other structural MRI studies of gray matter volume
loss in MCI [612,14]. As for the relationship between regional
alterations in DTI and GM atrophy, significant correlations were
limited to limbic structures and the corpus callosum. Although this
suggests that microstructural degeneration and macroscopic tissue
loss are correlated, the neural processes underlying the relationship
between cortical atrophy and decreased white matter integrity
remain unclear. Both retrograde and anterograde neuronal
degeneration [61] could be responsible for the finding. Longitudinal studies are needed to investigate the causal relationship
between white matter degradations and cortical atrophy in MCI
and AD.
The finding that DTI outperformed structural MRI GM
volume in the classification of MCI is surprising, because brain
atrophy is generally considered to be strongly associated with
cognitive impairments. Some MRI studies in MCI, using regional
brain volumes reported between 73% to 91% classification
accuracy [6269]. Thus, the 79.1% fitted accuracy under the
ROC curve for classifying MCI and healthy controls reported in
our study still appears to be consistent with the results in literature.
However, these previous studies employed mostly a single or a
small number of ROIs, such as the hippocampus or the entorhinal
cortex, achieving reasonably high sensitivity but lacking specificity
potentially because atrophy of the hippocampus and/or entorhinal
cortex are not features unique to MCI. By utilizing multiple brain
regions simultaneously for classifications, we may have captured
the heterogeneity of the MCI population better than evaluating
Figure 4. Classification accuracies based on DTI and GM

volume measures. Receiver operator characteristic curves of classifications of MCI and control subjects based on DTI and GM volume
measures used separately or together.
the splenium, fornix isthmus cingulum, and uncinate fasciculi, are

consistent with most other DTI studies with positive results
[16,17,24,28,29,31,33,54,55]. However, our results conflict with
some previous studies [15,18,25,5658] that found no significant
DTI alterations in some of these regions. The findings also do not
fully replicate our previous results [19] that showed significant FA
reduction in the parahippocampal cingulum of MCI subjects. The
inconsistent DTI findings might results partially from methodological differences between the studies. In particular, simply
artifacts of microscopic heterogeneity, such as crossing fibers, may
explain some of the inconsistencies. Moreover, heterogeneous
pathological underpinnings of study populations may also result
different DTI findings. For example, increasing evidences indicate
that cognitively normal elderly individuals show pathological
features of AD (i.e., neuritic plaques and neurofibrillary tangles). If
DTI is sensitive to capture already relevant pathologies in
cognitively normal group, the group differentiation with DTI
might be diminished. Nonetheless, our findings of abnormal DTI
Table 2. Group classifications based on either DTI or GM volume measures separately or used together.
Measure
Sensitivity (%) Specificity (%)Accuracy (%)
Fitted AUC (%) Cross-validated AUC (%)

(95% CI Lower)
median
(95% CI Upper)
FA
70.6
77.2
74.4
82.8
51.5
70.8
90.0
DR
66.7
78.8
73.5
83.5
58.0
78.5
95.0
GM volume
64.7
83.3
75.2
79.1
49.5
67.3
86.2
FA + GM volume
88.2
90.9
89.7
94.9
63.9
82.3a
96.2
DR + GM volume
82.3
86.4
84.6
92.6
61.9
81.2b
96.2
AUC area under a receiver operating characteristic curve. AUCs were tested using bootstrap (2000 boots):
Differences between modalities (i.e. FA + GM volume vs. GM volume) were significant at 95% confidence interval.
b
Differences between modalities (i.e. DR + GM volume vs. GM volume) were significant at 90% confidence interval.
a
61
Figure 5. Accuracies of predicting cognitive scores based on DTI and GM volume measures. Distribution of root mean square errors in
predicting MMSE and CVLT based on FA, DR or GM volume measures.
62
each region separately. In contrast to studies that measured

regional GM volumes, several DTI studies reported moderate
classification accuracies (6183%) of MCI [20,37,38,70] based on
a pre-selected single ROI. However, a complication in interpreting
previous DTI results is that the pre-selected single ROI were not in
consistent anatomical locations and the regional variations in DTI
across different brain locations were not considered. In this study
we attempted to take regional variations into account by using all
selected ROIs simultaneously for classifications. Thus the results
are less dependent on pre-selection of regions and potentially more
powerful. Our classification results, when considering multiple
regions, have shown similar cross-validated accuracies as that
reported in previous DTI studies. The findings that DTI
outperformed GM volumes over the cross-validated accuracies
imply that DTI measurement is as accurate as the GM volume
measurement in classifying MCI from healthy elderly controls.
Furthermore, another potential strength of the DTI measurement
is that DTI improves the diagnostic accuracies when using
together with GM volume measures, this finding reconfirms our
previous report and other reports in literature [19,37,38].
Our finding that DTI and GM volume measures were similarly
accurate in predicting cognitive performance in MCI has not been
previously reported. Generally, cognitive performance has been
associated with gray matter atrophy and less with white matter
alterations [7174]. In particular, measuring hippocampal volume
loss has often been used as predictor of memory deficits in MCI,
consistent with well-known role of the hippocampus for memory
consolidation [7577]. Mean diffusivity of the parahippocampi
[15] or the posterior cinguli [26] was reported predictive for
memory performance in previous DTI studies with small samples.
The abnormalities reported by DTI may reflect WM insults such
as myelin/axonal degeneration, oligodendrocytes death, reactive
gliosis, and small infarctions that may occur secondarily to volume
loss, which is known to be associated with memory deficits.
Consequently, it is reasonable to consider DTI as a possible
predictor to represent simultaneous though not identical neuropathological processes in addition to using GM volume. Although
we cannot determine in this cross-sectional study whether the

predictions based on DTI reflect primary and secondary effects,
our results imply that alterations in WM are similarly important
for predictions of cognitive performance compared to GM volume
loss. Moreover, by taking regional variations into account, we
reduced bias toward localized alterations that may not be
characteristic for MCI. We believe that our approach is more
reliable and also provides more interpretable results than testing a
few pre-determined regions separately. Our finding that DTI
achieved similar power as GM volumes in predicting cognitive
deficits highlights the potential value of DTI.
Our study has several limitations. It is possible that vascular
factors contributed to FA, DR, and GM volume differences
between MCI and controls [34,78], although we avoided brain
areas with visual WM hyperintensities when selecting ROIs. On
the other hand, some particular ROIs may include surrounding
voxels that contain GM, CSF and crossing fibers that may have
exaggerated group differences of DTI measures. Although in our
study DTI measures were limited in ROIs that contain voxels of
FA .0.2, there is no definition of a fiber boundary that enables
fully exclusion of partial volume. The potential influence of
vascular factors and partial volume effects on our results requires
further investigation but is beyond the scope of this study. Second,
the MCI group included not only subjects with memory
complaints (amnestic MCI), but also those with executive,
language or other deficits (non-amnestic MCI) as the dominant
clinical symptoms. Correspondingly, the group may also be
heterogeneous with respect to the pathological underpinnings of
the clinical features. Furthermore, the number of subjects in this
cohort, who converted to AD after 2-years clinical follow-up was
too small to determine whether the DTI and GM volume
alterations were linked to a conversion to AD with a reasonable
level of confidence. Studies with longer follow-ups are needed to
compare the predictive value of structural MRI and DTI for the
development of AD. A methodological limitation was that we did
not capture complete distributions of DTI and GM volume
variations across the whole brain. This may have biased our results
8

as well as reduced sensitivity as potential variations outside the

regions of interest were ignored. Although recently advances have
been made in processing all imaging voxels simultaneously, using
machine learning algorithms [79], the methodological complexity
of machine learning for MRI is not yet fully understood, especially
when multiple imaging variables are used together. As machine
learning methodologies in MRI become more robust, the
comparisons between DTI and structural MRI should be
repeated. Another methodological limitation is that DTI was
performed with only 6 diffusion gradient directions, while more
directions are known to stabilize tensor estimations. However, at
the time this study was started, a DTI protocol with more diffusion
directions was not available. It is therefore possible that our DTI
results are confounded by larger error margins than recent DTI
studies using more directions.
In conclusion, this study shows that DTI measures are as
accurate as measures of GM volumes in characterizing brain
alterations associated with cognitive impairment, and a combina-
tion of DTI and structural MRI measurements improves

classification accuracy.
Acknowledgments
The authors thank all the participants in this study. We also thank Mr.
Matthew Nordstrom, Mr. Frank Ezekiel for the assistance with image
processing, Mr. Derek Flenniken for clinical data organizing, and Mr.
Philip Insel for the help with statistics.
Conceived and designed the experiments: YZ NS MWW. Performed the
experiments: YZ NS MC LLC. Analyzed the data: YZ NS MC LLC.
Contributed reagents/materials/analysis tools: NS LLC TPF KY SCW
CM HJR BLM MWW. Wrote the paper: YZ NS. Designed the software
used in analysis: TPF. Results interpretation and manuscript editing: YZ
NS LLC KY SCW CM HJR BLM MWW.
References
1. Albert MS, DeKosky ST, Dickson D, Dubois B, Feldman HH, et al. (2011) The
diagnosis of mild cognitive impairment due to Alzheimers disease: recommendations from the National Institute on Aging-Alzheimers Association workgroups on diagnostic guidelines for Alzheimers disease. Alzheimers Dement 7:
270279.
2. Ramani A, Jensen JH, Helpern JA (2006) Quantitative MR imaging in
Alzheimer disease. Radiology 241: 2644.
3. Frisoni GB, Padovani A, Wahlund LO (2003) The diagnosis of Alzheimer
disease before it is Alzheimer dementia. Arch Neurol 60: 1023; author reply
10231024.
MRI and CSF biomarkers in normal, MCI, and AD subjects: diagnostic
discrimination and cognitive correlations. Neurology 73: 287293.
5. Sluimer JD, Bouwman FH, Vrenken H, Blankenstein MA, Barkhof F, et al.
(2010) Whole-brain atrophy rate and CSF biomarker levels in MCI and AD: a
longitudinal study. Neurobiol Aging 31: 758764.
6. Jack CR, Jr., Petersen RC, Xu YC, O9Brien PC, Smith GE, et al. (1999)
Prediction of AD with MRI-based hippocampal volume in mild cognitive
impairment. Neurology 52: 13971403.
7. Du AT, Schuff N, Amend D, Laakso MP, Hsu YY, et al. (2001) Magnetic
resonance imaging of the entorhinal cortex and hippocampus in mild cognitive
impairment and Alzheimers disease. J Neurol Neurosurg Psychiatry 71: 441
447.
8. Kantarci K, Xu Y, Shiung MM, O9Brien PC, Cha RH, et al. (2002)
Comparative diagnostic utility of different MR modalities in mild cognitive
impairment and Alzheimers disease. Dement Geriatr Cogn Disord 14: 198
207.
9. Grundman M, Jack CR, Jr., Petersen RC, Kim HT, Taylor C, et al. (2003)
Hippocampal volume is associated with memory but not monmemory cognitive
performance in patients with mild cognitive impairment. J Mol Neurosci 20:
241248.
10. Pennanen C, Kivipelto M, Tuomainen S, Hartikainen P, Hanninen T, et al.
(2004) Hippocampus and entorhinal cortex in mild cognitive impairment and
early AD. Neurobiol Aging 25: 303310.
11. Whitwell JL, Petersen RC, Negash S, Weigand SD, Kantarci K, et al. (2007)
Patterns of atrophy differ among specific subtypes of mild cognitive impairment.
Arch Neurol 64: 11301138.
12. Morra JH, Tu Z, Apostolova LG, Green AE, Avedissian C, et al. (2009)
Automated 3D mapping of hippocampal atrophy and its clinical correlates in
400 subjects with Alzheimers disease, mild cognitive impairment, and elderly
controls. Hum Brain Mapp 30: 27662788.
13. Petersen RC (2009) Early diagnosis of Alzheimers disease: is MCI too late? Curr
Alzheimer Res 6: 324330.
14. Frisoni GB, Fox NC, Jack CR, Jr., Scheltens P, Thompson PM (2010) The
clinical use of structural MRI in Alzheimer disease. Nat Rev Neurol 6: 6777.
15. Rogalski EJ, Murphy CM, deToledo-Morrell L, Shah RC, Moseley ME, et al.
(2009) Changes in parahippocampal white matter integrity in amnestic mild
cognitive impairment: a diffusion tensor imaging study. Behav Neurol 21: 5161.
16. Kiuchi K, Morikawa M, Taoka T, Nagashima T, Yamauchi T, et al. (2009)
Abnormalities of the uncinate fasciculus and posterior cingulate fasciculus in
mild cognitive impairment and early Alzheimers disease: a diffusion tensor
tractography study. Brain Res 1287: 184191.
17. Mielke MM, Kozauer NA, Chan KC, George M, Toroney J, et al. (2009)
Regionally-specific diffusion tensor imaging in mild cognitive impairment and
18. Stahl R, Dietrich O, Teipel SJ, Hampel H, Reiser MF, et al. (2007) White
matter damage in Alzheimer disease and mild cognitive impairment: assessment
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
with diffusion-tensor MR imaging and parallel imaging techniques. Radiology

243: 483492.
Zhang Y, Schuff N, Jahng GH, Bayne W, Mori S, et al. (2007) Diffusion tensor
imaging of cingulum fibers in mild cognitive impairment and Alzheimer disease.
Neurology 68: 1319.
Fellgiebel A, Wille P, Muller MJ, Winterer G, Scheurich A, et al. (2004)
Ultrastructural hippocampal and white matter alterations in mild cognitive
impairment: a diffusion tensor imaging study. Dement Geriatr Cogn Disord 18:
101108.
Bozoki AC, Korolev IO, Davis NC, Hoisington LA, Berger KL (2011)
Disruption of limbic white matter pathways in mild cognitive impairment and
Alzheimers disease: A DTI/FDG-PET Study. Hum Brain Mapp 33: 1792
1802.
Oishi K, Akhter K, Mielke M, Ceritoglu C, Zhang J, et al. (2011) Multi-modal
MRI analysis with disease-specific spatial filtering: initial testing to predict mild
cognitive impairment patients who convert to Alzheimers disease. Front Neurol
2: 54.
Teipel SJ, Meindl T, Grinberg L, Grothe M, Cantero JL, et al. (2010) The
cholinergic system in mild cognitive impairment and Alzheimers disease: an in
vivo MRI and DTI study. Hum Brain Mapp 32: 13491362.
Zhuang L, Wen W, Zhu W, Trollor J, Kochan N, et al. (2010) White matter
integrity in mild cognitive impairment: a tract-based spatial statistics study.
Walhovd KB, Fjell AM, Amlien I, Grambaite R, Stenset V, et al. (2009)
Multimodal imaging in mild cognitive impairment: Metabolism, morphometry
and diffusion of the temporal-parietal memory network. Neuroimage 45: 215
223.
Bozzali M, Giulietti G, Basile B, Serra L, Spano B, et al. (2011) Damage to the
cingulum contributes to Alzheimers disease pathophysiology by deafferentation
mechanism. Hum Brain Mapp 33: 12951308.
Grambaite R, Selnes P, Reinvang I, Aarsland D, Hessen E, et al. (2011)
Executive dysfunction in mild cognitive impairment is associated with changes in
frontal and cingulate white matter tracts. J Alzheimers Dis 27: 453462.
Pievani M, Agosta F, Pagani E, Canu E, Sala S, et al. (2010) Assessment of white
matter tract damage in mild cognitive impairment and Alzheimers disease.
Hum Brain Mapp 31: 18621875.
Chua TC, Wen W, Chen X, Kochan N, Slavin MJ, et al. (2009) Diffusion tensor
imaging of the posterior cingulate is a useful biomarker of mild cognitive
impairment. Am J Geriatr Psychiatry 17: 602613.
Fellgiebel A, Muller MJ, Wille P, Dellani PR, Scheurich A, et al. (2005) Colorcoded diffusion-tensor-imaging of posterior cingulate fiber tracts in mild
cognitive impairment. Neurobiol Aging 26: 11931198.
Choo IH, Lee DY, Oh JS, Lee JS, Lee DS, et al. (2010) Posterior cingulate
cortex atrophy and regional cingulum disruption in mild cognitive impairment
and Alzheimers disease. Neurobiol Aging 31: 772779.
Sexton CE, Kalu UG, Filippini N, Mackay CE, Ebmeier KP (2010) A metaanalysis of diffusion tensor imaging in mild cognitive impairment and
Alzheimers disease. Neurobiol Aging 32: 2322 e23252318.
Ukmar M, Makuc E, Onor ML, Garbin G, Trevisiol M, et al. (2008) Evaluation
of white matter damage in patients with Alzheimers disease and in patients with
mild cognitive impairment by using diffusion tensor imaging. Radiol Med 113:
915922.
Delano-Wood L, Bondi MW, Jak AJ, Horne NR, Schweinsburg BC, et al. (2010)
Stroke risk modifies regional white matter differences in mild cognitive
impairment. Neurobiol Aging 31: 17211731.
63
35. Song SK, Yoshino J, Le TQ, Lin SJ, Sun SW, et al. (2005) Demyelination
increases radial diffusivity in corpus callosum of mouse brain. Neuroimage 26:
132140.
36. Acosta-Cabronero J, Williams GB, Pengas G, Nestor PJ (2010) Absolute
diffusivities define the landscape of white matter degeneration in Alzheimers
37. Wang L, Goldstein FC, Veledar E, Levey AI, Lah JJ, et al. (2009) Alterations in
cortical thickness and white matter integrity in mild cognitive impairment
measured by whole-brain cortical thickness mapping and diffusion tensor
imaging. AJNR Am J Neuroradiol 30: 893899.
38. Cui Y, Wen W, Lipnicki DM, Beg MF, Jin JS, et al. (2011) Automated detection
of amnestic mild cognitive impairment in community-dwelling elderly adults: a
combined spatial atrophy and white matter alteration approach. Neuroimage
59: 12091217.
39. Folstein MF, Folstein SE, McHugh PR (1975) Mini-mental state. A practical
Res 12: 189198.
40. Morris JC (1993) The Clinical Dementia Rating (CDR): current version and
41. Delis DC, Kaplan E, Kramer JH (2001) Delis-Kaplan Executive Function
System. San Antonio: The Psychological Corporation.
42. Delis DC, Kramer JH, Kaplan E, Ober BA (2000) California Verbal Learning
Test, 2nd Edition. San Antonio, TX: The Psychological Corporation.
43. Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, et al. (2002)
Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn
Reson Med 47: 12021210.
44. Fischl B, Salat DH, Busa E, Albert M, Dieterich M, et al. (2002) Whole brain
segmentation: automated labeling of neuroanatomical structures in the human
brain. Neuron 33: 341355.
45. Fischl B, van der Kouwe A, Destrieux C, Halgren E, Segonne F, et al. (2004)
Automatically parcellating the human cerebral cortex. Cereb Cortex 14: 1122.
46. Callen DJ, Black SE, Gao F, Caldwell CB, Szalai JP (2001) Beyond the
hippocampus: MRI volumetry confirms widespread limbic atrophy in AD.
Neurology 57: 16691674.
47. Jones BF, Barnes J, Uylings HB, Fox NC, Frost C, et al. (2006) Differential
regional atrophy of the cingulate gyrus in Alzheimer disease: a volumetric MRI
study. Cereb Cortex 16: 17011708.
48. Jacobs HI, Van Boxtel MP, Jolles J, Verhey FR, Uylings HB (2011) Parietal
cortex matters in Alzheimers disease: an overview of structural, functional and
metabolic findings. Neurosci Biobehav Rev 36: 297309.
49. Tao R, Fletcher PT, Gerber S, Whitaker RT (2009) A variational image-based
approach to the correction of susceptibility artifacts in the alignment of diffusion
weighted and structural MRI. Inf Process Med Imaging 21: 664675.
50. Chua TC, Wen W, Slavin MJ, Sachdev PS (2008) Diffusion tensor imaging in
mild cognitive impairment and Alzheimers disease: a review. Curr Opin Neurol
21: 8392.
51. Benjamini Y, Drai D, Elmer G, Kafkafi N, Golani I (2001) Controlling the false
discovery rate in behavior genetics research. Behavioural Brain Research 125:
279284.
52. Hastie T, Tibshirani RJ, Friedman J (2009) The Elements of Statistical
Learning: Data Mining, Inference, and Prediction.: Springer.
53. Clerx L, Visser PJ, Verhey F, Aalten P (2012) New MRI markers for Alzheimers
disease: a meta-analysis of diffusion tensor imaging and a comparison with
medial temporal lobe measurements. J Alzheimers Dis 29: 405429.
54. Cho H, Yang DW, Shon YM, Kim BS, Kim YI, et al. (2008) Abnormal integrity
of corticocortical tracts in mild cognitive impairment: a diffusion tensor imaging
study. J Korean Med Sci 23: 477483.
55. Ringman JM, O9Neill J, Geschwind D, Medina L, Apostolova LG, et al. (2007)
Diffusion tensor imaging in preclinical and presymptomatic carriers of familial
Alzheimers disease mutations. Brain 130: 17671776.
56. Chen TF, Lin CC, Chen YF, Liu HM, Hua MS, et al. (2009) Diffusion tensor
changes in patients with amnesic mild cognitive impairment and various
dementias. Psychiatry Res 173: 1521.
57. Bai F, Zhang Z, Watson DR, Yu H, Shi Y, et al. (2009) Abnormal integrity of
association fiber tracts in amnestic mild cognitive impairment. J Neurol Sci 278:
102106.
58. Rose SE, McMahon KL, Janke AL, O9Dowd B, de Zubicaray G, et al. (2006)
Diffusion indices on magnetic resonance imaging and neuropsychological
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
performance in amnestic mild cognitive impairment. J Neurol Neurosurg

Hoeft F, Barnea-Goraly N, Haas BW, Golarai G, Ng D, et al. (2007) More is not
always better: increased fractional anisotropy of superior longitudinal fasciculus
associated with poor visuospatial abilities in Williams syndrome. J Neurosci 27:
1196011965.
Douaud G, Jbabdi S, Behrens TE, Menke RA, Gass A, et al. (2011) DTI
measures in crossing-fibre areas: increased diffusion anisotropy reveals early
white matter alteration in MCI and mild Alzheimers disease. Neuroimage 55:
880890.
Pearson RC, Powell TP (1987) Anterograde vs. retrograde degeneration of the
nucleus basalis medialis in Alzheimers disease. Journal of Neural TransmissionSupplementum 24: 139146.
Xu Y, Jack CR Jr., O9Brien PC, Kokmen E, Smith GE, et al. (2000) Usefulness
of MRI measures of entorhinal cortex versus hippocampus in AD. Neurology 54:
17601767.
De Santi S, de Leon MJ, Rusinek H, Convit A, Tarshish CY, et al. (2001)
Hippocampal formation glucose metabolism and volume losses in MCI and AD.
Wolf H, Grunwald M, Kruggel F, Riedel-Heller SG, Angerhofer S, et al. (2001)
Hippocampal volume discriminates between normal cognition; questionable and
mild dementia in the elderly. Neurobiol Aging 22: 177186.
Bottino CM, Castro CC, Gomes RL, Buchpiguel CA, Marchetti RL, et al.
(2002) Volumetric MRI measurements can differentiate Alzheimers disease,
mild cognitive impairment, and normal aging. Int Psychogeriatr 14: 5972.
Slavin MJ, Sandstrom CK, Tran TT, Doraiswamy PM, Petrella JR (2007)
Hippocampal volume and the Mini-Mental State Examination in the diagnosis
of amnestic mild cognitive impairment. AJR Am J Roentgenol 188: 14041410.
Colliot O, Chetelat G, Chupin M, Desgranges B, Magnin B, et al. (2008)
Discrimination between Alzheimer disease, mild cognitive impairment, and
normal aging by using automated segmentation of the hippocampus. Radiology
248: 194201.
Jauhiainen AM, Pihlajamaki M, Tervo S, Niskanen E, Tanila H, et al. (2009)
Discriminating accuracy of medial temporal lobe volumetry and fMRI in mild
cognitive impairment. Hippocampus 19: 166175.
Desikan RS, Cabral HJ, Hess CP, Dillon WP, Glastonbury CM, et al. (2009)
Automated MRI measures identify individuals with mild cognitive impairment
and Alzheimers disease. Brain 132: 20482057.
Muller MJ, Greverus D, Weibrich C, Dellani PR, Scheurich A, et al. (2007)
Diagnostic utility of hippocampal size and mean diffusivity in amnestic MCI.
Grundman M, Sencakova D, Jack CR Jr., Petersen RC, Kim HT, et al. (2002)
Brain MRI hippocampal volume and prediction of clinical status in a mild
cognitive impairment trial. J Mol Neurosci 19: 2327.
Loewenstein DA, Acevedo A, Potter E, Schinka JA, Raj A, et al. (2009) Severity
of medial temporal atrophy and amnestic mild cognitive impairment: selecting
type and number of memory tests. Am J Geriatr Psychiatry 17: 10501058.
Ystad MA, Lundervold AJ, Wehling E, Espeseth T, Rootwelt H, et al. (2009)
Hippocampal volumes are important predictors for memory function in elderly
women. BMC Med Imaging 9: 17.
Walhovd KB, Fjell AM, Dale AM, McEvoy LK, Brewer J, et al. (2010) Multimodal imaging predicts memory performance in normal aging and cognitive
decline. Neurobiol Aging 31: 11071121.
Edison P, Archer HA, Hinz R, Hammers A, Pavese N, et al. (2007) Amyloid,
hypometabolism, and cognition in Alzheimer disease: an [11C]PIB and
[18F]FDG PET study. Neurology 68: 501508.
Mosconi L, Brys M, Glodzik-Sobanska L, De Santi S, Rusinek H, et al. (2007)
Early detection of Alzheimers disease using neuroimaging. Exp Gerontol 42:
129138.
Petersen RC, Jack CR Jr., Xu YC, Waring SC, O9Brien PC, et al. (2000)
Memory and MRI-based hippocampal volumes in aging and AD. Neurology 54:
581587.
Lee DY, Fletcher E, Martinez O, Ortega M, Zozulya N, et al. (2009) Regional
pattern of white matter microstructural changes in normal aging, MCI, and AD.
Neurology 73: 17221728.
Davatzikos C, Ruparel K, Fan Y, Shen DG, Acharyya M, et al. (2005)
Classifying spatial patterns of brain activity with machine learning methods:
application to lie detection. Neuroimage 28: 663668.
64
10
Neuroimaging
biomarkers:
PET
65
Diagnosed Mild Cognitive Impairment Due to

Alzheimers Disease with PET Biomarkers of Beta
Amyloid and Neuronal Dysfunction
Shizuo Hatashita*, Hidetomo Yamasaki
Neurology, Shonan-Atsugi Hospital, Atsugi, Japan
Abstract
The aim of this study is to identify mild cognitive impairment (MCI) due to Alzheimers disease (AD) using amyloid imaging
of beta amyloid (Ab) deposition and FDG imaging of reflecting neuronal dysfunction as PET biomarkers. Sixty-eight MCI
patients underwent cognitive testing, [11C]-PIB PET and [18F]-FDG PET at baseline and follow-up. Regions of interest were
defined on co-registered MRI. PIB distribution volume ratio (DVR) was calculated using Logan graphical analysis, and the
standardized uptake value ratio (SUVR) on the same regions was used as quantitative analysis for [18F]-FDG. Thirty (44.1%)
of all 68 MCI patients converted to AD over 19.267.1 months. The annual rate of MCI conversion was 23.4%. A positive Ab
PET biomarker significantly identified MCI due to AD in individual MCI subjects with a sensitivity (SS) of 96.6% and specificity
(SP) of 42.1%. The positive predictive value (PPV) was 56.8%. A positive Ab biomarker in APOE e4/4 carriers distinguished
with a SS of 100%. In individual MCI subjects who had a prominent impairment in episodic memory and aged older than 75
years, an Ab biomarker identified MCI due to AD with a greater SS of 100%, SP of 66.6% and PPV of 80%, compared to FDG
biomarker alone or both PET biomarkers combined. In contrast, when assessed in precuneus, both Ab and FDG biomarkers
had the greatest level of certainty for MCI due to AD with a PPV of 87.8%. The Ab PET biomarker primarily defines MCI due
to AD in individual MCI subjects. Furthermore, combined FDG biomarker in a cortical region of precuneus provides an
added diagnostic value in predicting AD over a short period.
Citation: Hatashita S, Yamasaki H (2013) Diagnosed Mild Cognitive Impairment Due to Alzheimers Disease with PET Biomarkers of Beta Amyloid and Neuronal
Dysfunction. PLoS ONE 8(6): e66877. doi:10.1371/journal.pone.0066877
Editor: Stephen D Ginsberg, Nathan Kline Institute and New York University School of Medicine, United States of America
Received February 8, 2013; Accepted May 10, 2013; Published June 14, 2013
Copyright: 2013 Hatashita and Yamasaki. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The authors have no support or funding to report.
Competing Interests: GE Healthcare Ltd. (Little Chalfont, Buckinghamshire, HP7 9NA, United Kingdom) entered into the sublicense agreement with the
Shonan-Atsugi Hospital to allow the use of the [11C]-PIB PET imaging. There are no further patents, products in development or marketed products to declare.
This does not alter the authors adherence to all the PLOS ONE policies on sharing data and materials, as detailed online in the guide for authors.
beta-amyloid (Ab) 42, while other biomarkers reflect downstream

neuronal injury associated with AD, including CSF tau (both total
tau and phosphorylated tau), fluorodeoxyglucose (FDG)-PET
imaging and MRI. The evidence of both Ab deposition and
neuronal injury together provide the greatest probability that the
AD pathophysiological processes are present. A number of studies
have compared the predictive value of two or more biomarkers at
a time but findings have been inconsistent [46]. Some aspects of
the clinical research criteria may be revised, as these criteria are
put into practice and new findings become available.
The PET amyloid imaging with a tracer of [11C]-labeled
Pittsburgh Compound-B ([11C]-PIB:N-methyl-[11C]2-[49-methylaminophenyl]-6-hydroxybenzothiazole), which has a high affinity
for fibrillar Ab, is a reliable biomarker of underlying AD pathology
[7,8]. On the other hand, a FDG-PET imaging for cerebral
glucose metabolism, reflecting downstream neuronal injury,
provides evidence about cognitive function and progression that
can not be provided by amyloid PET imaging. We have recently
demonstrated that a diagnostic framework with an Ab deposition
by [11C]-PIB PET, in different clinical stages of AD, allows for an
earlier and more specific AD diagnosis [9]. In addition, our
preliminary study with FDG-PET imaging showed that cerebral
glucose metabolism starts to decrease in prodromal AD, and
Introduction
Mild cognitive impairment (MCI) has a multitude of causes,
including predementia of Alzheimers disease (AD) and non-AD as
well as depression and various physical disorders. A recent study
has shown that subjects with MCI are at increased risk of
developing AD and that their overall rate of progression to AD is
typically 10%215% per year [1]. Identifying subjects with MCI
who are most likely to decline in cognition over time is a major
focus in AD research. Revised research criteria for the diagnosis of
AD have been proposed, and a framework has been developed to
capture the earliest stage of AD [2]. Recently, the National
Institute on Aging (NIA) - Alzheimers Association working group
has proposed the diagnostic criteria for the symptomatic predementia phase of AD, referred to a MCI due to AD [3]. The
MCI due to AD could be used to identify individuals with AD
pathophysiological processes as the primary cause of their
progressive cognitive dysfunction.
The clinical research diagnosis of MCI due to AD is based on
the Core Clinical Criteria used in all clinical settings without the
need of highly specialized tests and the Clinical Research Criteria
with the use of biomarkers. Some biomarker measurements
directly reflect the pathology of AD, including positron emission
tomography (PET) amyloid imaging and cerebrospinal fluid (CSF)
66

MCI Due to AD with Ab and FDG PET Biomarkers
Fifteen-minute static FDG-PET scans were acquired with the

same camera after a 45-minute uptake period, and the image was
reconstructed using the same image reconstruction techniques.
greater decrease correlates with greater cognitive impairment

along the continuum from prodromal to AD dementia.
The present study was conducted to define the MCI due to AD
using PET amyloid imaging of Ab deposition and FGD-PET
imaging of neuronal dysfunction as a biomarker. We sought to
determine whether episodic memory impairment, age and
apolipoprotein-E (APOE) genotype have an effect on progression
to AD dementia.
Image analysis
All subjects underwent T1-weighed MRI (1.5 T) for coregistration with the PET images. MRI-based correction of PET
data for partial volume effects was carried out using the PMOD
software (PMOD Technologies Ltd., Adliswil, Switzerland).
Regions of interest were manually drawn on the co-registered
MR image, including the following cortical regions: lateral
temporal cortex, medial temporal cortex, frontal cortex, occipital
cortex, parietal cortex, sensory motor cortex, anterior cingulate
gyrus, posterior cingulate gyrus, precuneus cortex and cerebellar
cortex.

Subjects
Sixty-eight Japanese participants were recruited from our
memory clinic. All were between 50 and 90 years of age. They
underwent neurological and neuropsychological assessment, and
neuroimaging at baseline and once more after 1224 months.
Global cognitive status was assessed with the Mini-Mental-State
Examination (MMSE) [10] and severity of dementia was rated on
the Clinical Dementia Rating (CDR) scale [11]. The CDR sum of
boxes (CDR SB) score was a simple sum of the score obtained in
each of the 6 rated domains. A memory measure of immediate and
delayed recall of a paragraph from the Wechsler Memory ScaleRevised (WMS-R) Logical Memory II was carried out as a simple
episodic memory test [12]. The apolipoprotein E (APOE)
genotype was determined from venous blood samples.
All 68 MCI patients met the Core Clinical Criteria for MCI,
proposed by the NIA-Alzheimers Association workgroup [3],
including concern about a change in cognition, impairment in one
or more cognitive domains, preservation of independence in
functional abilities, and no dementia. Other systemic or brain
diseases that could account for the decline in cognition, including
degenerative, vascular, depressive, traumatic, medical comorbidities, or mixed disease were excluded. The MMSE score was
greater than or equal to 24, and global CDR score was at least 0.5
in the memory domain. For the episodic memory measure,
delayed recall of a paragraph from the WMS-R Logical Memory
II (maximum score 25) was used with the following cutoff scores
based on education: these scores were less than 8 for subject with
$16 years of education, 4 for subjects with 815 years of
education, and 2 for subjects with 027 years of education [12].
AD diagnosis was based on the criteria of the National Institute of
Neurological and Communicative Disorders and Stroke-Alzheimers Disease and Related Disorders Association (NINCDSADRDA) [13].
Each subject or their caregiver provided written informed
consent for participation. The study protocol was approved by the
Institutional Review Board of the Mirai Iryo Research Center Inc
(Tokyo, Japan).
Data management
Levels of PIB retention were determined by the distribution
volume ratio (DVR) with Logan graphical analysis for 35 to 60
minutes with cerebellar gray matter as reference [14]. Regional
DVR values of each cortical region and the cortical DVR value of
the whole cortical regions were calculated. DVR PET images were
created for visual inspection with a rainbow color scale. Scans of
all subjects were visually assessed with 2 readers as amyloidpositive or amyloid-negative for cortical PIB retention. Amyloidpositive images showed greatly increased PIB retention in the
cortex compared to white matter and a typical PIB pattern of
cortical amyloid deposition, whereas amyloid-negative images
showed no PIB retention in any of the cortical regions.
The same co-registration method was applied for quantification
of [18F]-FDG. A standardized uptake value (SUV) of the same
region was obtained and subsequently normalized to the cerebellar
cortex as reference. Glucose metabolism was referred to as the
SUV ratio (SUVR).
Statistical analysis
Group differences were evaluated with analysis of variance
(ANOVA), followed by Bonferroni post hoc tests to assess the
significance. Analysis correlation between DVR values, FDG
SUVR values, age, MMSE scores, CDR SB scores and WMS-R
recall scores yielded Pearsons product moment correlation
coefficient (r). Categorical variables were examined with Fishers
exact test. Paired t-tests were used to study changes between
baseline and follow-up data. Results were considered significant at
p,0.05. Data are presented as means 6 standard deviations (SD).
Statistical analyses were performed with Statcel 3 software (OMS
Inc. Japan).
Results
PET imaging
Clinical data and cognitive function
All PET scans were performed on the same day as the cognitive
testing, using a Siemens ECAT ACCEL scanner in 3-dimensional
scanning mode, providing 63 contiguous 2.46-mm slices with a
5.6-mm transaxial and a 5.4-mm axial resolution. All imaging data
were reconstructed into a 1286128 matrix. The amyloid PET
imaging was accomplished with the radiotracer [11C]-PIB, as
previously described in detail [9]. Briefly, [11C]-PIB was produced
in our PET center using the one-step [11C] methyl triflate
approach. [11C]-PIB was injected intravenously as a bolus with a
mean dose of 561.5611.2 MBq (n = 68). Dynamic PET scanning
was performed for 60 minutes according to a predetermined
protocol. Sixty minutes following the completion of [11C]-PIB
image, the subjects were injected intravenously with 249.9628.8
MBq (n = 68) of [18F]-FDG and remained in a dark, quiet room.
Thirty (44.1%) of the 68 patients with MCI converted to AD

during the follow-up of 19.267.1 months (converters), while the
remaining 38 (55.9%) did not progress to AD (stable patients).
Twelve (50%) of 24 MCI patients aged 75289 years converted to
AD, compared to 18 (40.9%) of 44 MCI patients aged 50274
years. Fourteen (50%) of 28 APOE e4 carriers with MCI and 16
(40%) of 40 APOE e4 non-carriers converted to AD. In particular,
5 (83.3%) of 6 APOE e4/4 carriers with MCI converted to AD.
Overall rate of MCI progression to AD was 23.4% per year.
The demographic characteristics of MCI converters and stable
patients at baseline and follow-up are summarized in Table 1. The
MMSE and CDR SB scores in MCI converters at baseline did not
differ from those in stable patients. The mean MMSE score of
2
67
MCI converters decreased to 22.862.5 at follow-up and their

CDR SB score increased to 2.561.0. The mean MMSE score of
MCI converters decreased approximately 4 points whereas that of
stable patients remained unchanged. There were no significant
differences in age or sex between MCI converters and stable
patients. None of the patients with MCI reverted to a normal
status or converted to a non-AD dementia.
For the memory measure of WMS-R Logical Memory II
Immediate and Delayed Recall, the mean delayed paragraph
recall score at baseline was 1.762.0 for converters, which were
significantly lower than the mean score for stable patients despite
their having approximately the same level of education (Table 1).
However, there was no significant difference in the mean
immediate paragraph recall scores between MCI converters and
stable patients.
Ab deposition
Fifty-one (75%) of the 68 MCI patients had marked PIB
retention in the frontal, parietal and lateral temporal cortical
regions as well as the cingulate gyrus and precuneus, showing the
typical AD pattern at baseline (Figure 1). Of 51 MCI patients who
had Ab deposition, 29 (56.8%) patients converted to AD during
the follow-up period while the remaining 22 patients did not
convert despite similar Ab deposition. In contrast, 16 of 17 MCI
patients who had no PIB retention in any cortical region were
stable and only 1 MCI patient converted to AD.
Cortical PIB DVR values for MCI converters and stable
patients are presented in Figure 2. At baseline, the mean value of
cortical DVR in amyloid-positive converters (2.0260.30, n = 29,
p,0.01) was significantly higher than that in amyloid-negative
stable patients (1.2660.16, n = 16). In the amyloid-positive stable
patients, in addition, the mean cortical DVR value (2.0560.37,
n = 22) was similar to amyloid-positive converters. Also, there were
no significant differences in the regional DVR values of any
cortical region between amyloid-positive converters and stable
patients. The mean value of cortical DVR in amyloid-positive
converters at follow-up (2.1160.37, n = 29) was not significantly
different from baseline (Figure 2). There were no significant
changes in cortical DVR values between baseline and follow-up in
amyloid-positive or amyloidnegative stable patients.
Figure 1. PIB-PET DVR images from 4 representative MCI

converters and stable patients with (+) and without amyloid
deposition (2) at baseline.
Glucose metabolism
Fifty-seven (83.8%) of 68 MCI patients had reduced glucose
metabolism (FDG SUVR #0.99) in whole cortical regions at
baseline, 28 patients (49.1%) of whom progressed to AD during
the follow-up period. A mean FDG SUVR value for the whole
cortical regions in converters was 0.9360.06 (n = 30, p,0.01) at
baseline, significantly different from stable patients (0.9860.06,
n = 38). In cortical regions of the lateral temporal cortex, parietal
cortex and precuneus, in particular, MCI converters had
significantly lower regional FDG SUVR values at baseline
compared to stable patients (Figure 3). At follow-up, MCI
converters did not have significant decreases in mean values of
FDG SUVR in whole cortical regions or each cortical region.
Table 1. Demographic Characteristics of Patients With MCI.
Baseline
68
Follow up
Characteristic
Converter
Stable
Converter
Stable
study popu
30
38
30
38
Education (yr)
12.262.4
12.062.0
12.262.4
12.062.0
MMSE
26.561.5
27.361.6
22.862.5*
27.361.6
global CDR
0.5
0.5
0.660.2
0.5
CDR SB
0.860.2
0.860.3
2.561.0*
0.760.4
Immediate Rec.
4.962.7
6.963.1
4.062.8
7.263.9
Delayed Rec.
1.762.0*
4.663.4
1.562.3
4.964.3
APOE e4/4, e4/3
14 (46.6%)
14 (36.8%)
14 (46.6%)
14 (36.8%)
Figure 2. The cortical PIB DVR values in PIB-positive MCI

converters (closed circles, n = 29), PIB-negative MCI converter
(open circles, n = 1), PIB-positive stable MCI patients (closed
triangles, n = 22), and PIB-negative stable MCI patients (open
triangles, n = 16) at baseline and follow-up. Data are presented as
mean 6 SD. *Statistically significant difference from the PIB-negative
stable MCI patients by multiple comparisons post hoc tests (p,0.01).
MCI: mild cognitive impairment, MMSE: Mini-Mental State Examination, CDR:

Clinical Dementia Rating, CDR SB: Clinical Dementia Rating sum of boxes score,
study popu: number of patients in the study population, yr: years. Rec: WMS-R
recall scores, APOE: apolipoprotein E, Data are presented as means 6 SD,
* Statistically significant difference from stable patients by multiple
comparisons post hoc tests (p,0.05).

Figure 3. Regional FDG SUVR values in the lateral temporal

cortex, frontal cortex, precuneus and parietal cortex of MCI
converters (closed circles, n = 30) and stable patients (open
circles, n = 38) at baseline. Data are presented as means 6 SD.
*Statistically significant difference from the stable patients by multiple
comparisons post hoc tests (p,0.05).
Figure 4. Scatter plot of the relationship between cortical PIB

DVR and FDG SUVR values in individual MCI converters (closed
circles, n = 30) and stable patients (open circles, n = 38). The
horizontal dotted line indicates the greatest value of PIB DVR
($1.49). The vertical dotted line indicates the lower value of FDG SUVR
(#0.99).
Comparison between Ab deposition and glucose

metabolism
sensitivity of 100% and a specificity of 0%. Combining Ab and

FDG PET biomarkers did not improve their sensitivity and
specificity in individual APOE e4 carriers and non-carriers.
When individual MCI patients, aged 75289 years, had lower
WMS-R Logical Memory (delayed recall score #4), a positive Ab
PET biomarker had greater sensitivity 100% and specificity 66.6%
at identifying MCI due to AD (p,0.01), and the positive
predictive value was 80%. However, combining positive Ab and
FDG PET biomarkers did not improve the discrimination of MCI
due to AD compared to positive Ab PET biomarker alone. In
contrast, when assessed in a cortical region of precuneus, a
combination of positive Ab and FDG PET biomarkers discriminated MCI due to AD from individual MCI subjects with the
Individual cortical FDG SUVR values were not correlated

significantly with PIB DVR values in all MCI patients at baseline
(r = 0.22, p = 0.06, n = 68). Also, there was also no significant
correlation between regional FDG and DVR values in the lateral
temporal cortex, frontal cortex, parietal cortex or precuneus.
Diagnostic value of PET biomarkers

The relationships between individual cortical PIB DVR and
FDG SUVR values for all 68 MCI patients are shown in Figure 4.
Individual PIB DVR values for 29 (96.6%) of 30 converters and 22
(57.8%) of 38 stable patients exceeded the lowest value of amyloidpositive subjects (DVR $1.49), which is used as a positive Ab PET
biomarker of Ab deposition. On the other hand, individual FDG
SUVR values were below the lower value (FDG SUVR #0.99),
which is used as positive FDG PET biomarker, for 28 (93.3%) of
30 converters and 29 (76.3%) of 38 stable patients. Positive or
negative value fell within the reliable range, as previously
demonstrated [9], and standardization of biomarkers for both
PET amyloid imaging and FDG-PET imaging was possible in our
clinical setting.
We used a dichotomous test using chi-square analysis to assess
the sensitivity (SS) and specificity (SP) of positive Ab PET and
FDG PET biomarkers for MCI due to AD (Table 2). A positive Ab
PET biomarker significantly identified MCI due to AD in
individual MCI subjects with a sensitivity of 96.6% and a
specificity of 42.1%, compared to negative Ab PET biomarker
(p,0.01). The positive predictive value (PPV) was 56.8%. A
positive FDG PET biomarker had a sensitivity of 93.3% and a
specificity of 23.6%. Combined positive Ab and FDG biomarkers
did not significantly discriminate MCI due to AD in individual
MCI subjects.
In individual MCI subjects with APOE e4, a positive Ab PET
biomarker significantly discriminated MCI due to AD
(SS = 100%, SP = 28.5%) in addition to individual APOE e4
non-carriers (SS = 93.7%, SP = 50%), compared to negative
biomarker. In particular, a positive Ab PET biomarker in the
individual MCI subjects with APOE e4/4 distinguished with a
Table 2. Sensitivity and specificity of positive Ab and/or FDG

PET biomarkers for MCI due to AD in individual MCI patients.
Biomarker
SS
SP
PPV
NPV
aged 5089
Ab PET
96.6
42.1
56.8
94.1
FDG PET
93.3
23.6
49.1
81.8
Ab+ FDG PET
96.5
21.7
60.8
83.3
aged $75 years, lower delayed recall

Ab PET
100
66.6
80.0
100
FDG PET
91.6
44.4
68.7
80.0
Ab+ FDG PET
100
60.0
84.6
100
APOE e4 carriers
Ab PET
100
28.5
58.3
100
FDG PET
100
21.4
56.0
100
Ab+ FDG PET
100
18.1
60.8
100
SS: sensitivity, SP: specificity, PPV: positive predictive value, NPV: negative
predictive value, Ab: b-amyloid, MCI: mild cognitive impairment, APOE:
apolipoprotein E, Data are presented as percentages.
69
greatest sensitivity of 96.6% and specificity of 73.3% (P,0.01).

The combination had a higher positive predictive value of 87.8%.
values for subject categorization, study design, and statistical

analysis. To determine MCI due to AD, the biomarker in
diagnosis reflecting Ab and downstream neuronal injury has been
proposed to be most important [3].
In the present study, we applied PET biomarkers to diagnose
MCI due to AD in individual MCI subjects. An Ab PET
biomarker identified MCI due to AD with a sensitivity of 96.6%
and a specificity of 42.1%. In individual APOE e4/4 carriers with
MCI, in particular, the Ab PET biomarker distinguished MCI due
to AD with greater certainty. Furthermore, when MCI subjects
had a prominent impairment in episodic memory and aged older
than 75 years, an Ab PET biomarker provided a further increased
level of certainty with a sensitivity of 100% and a specificity of
66.6%, compared to FDG PET biomarker alone or both PET
biomarkers combined. A recent meta-analysis has shown that
sensitivity and specificity of PIB-PET for predicting clinical
progression to AD range from 83.3100% and 41.776.8%,
respectively [20]. These findings indicate that the sensitivity of an
Ab PET biomarker is high in the prediction of conversion to AD
while the specificity is low. The Ab PET biomarker is primarily
required for diagnosing MCI due to AD in individual MCI
subjects.
Furthermore, in this study, a positive FDG PET alone or
combined positive Ab and FDG biomarkers did not significantly
discriminate MCI due to AD in individual MCI subjects. Although
a FDG PET biomarker has been associated with the neuropathology of AD, it is not specific for AD. In contrast, when assessed
in the precuneus with regional hypometabolism, both Ab and
FDG PET biomarkers had a greater level of certainty for MCI due
to AD with the greatest sensitivity of 96.6% and specificity of
73.3%. The combination had the highest positive predictive value
of 87.8%. We suggests that most individuals with a diagnosis of
MCI due to AD, using Ab and FDG PET biomarkers in a cortical
region of precuneus, progress to AD even within a short period.
The MCI due to AD could be more certainly diagnosed in
individual MCI subjects, based on core clinical criteria, with the
conjugation of biomarkers of PET amyloid imaging and FDGPET imaging. The clinical research diagnosis of MCI due to AD
would have the greatest potential benefit for assessing risk level in
single patients before the onset of the disease, providing a basis for
earlier therapeutic and pharmacologic intervention, patient
counseling, and the design of clinical trials.
Discussion
We demonstrated that 44.1% of 68 MCI patients converted to
AD during a follow-up of 19.267.1 months and the rate of MCI
conversion was 23.4% per year. A recent study has reported that
48% of 31 subjects with MCI converted to AD during a follow-up
period of 2.9 years [15]. The Alzheimers Disease Neuroimaging
Initiative has reported that 32.9% of 85 single-domain or multipledomain amnestic MCI patients converted to AD during a followup period of 1.960.4 years (an annual rate of 17.2%) [1]. Thus,
conversion rates vary greatly depending on the criteria applied, the
nature of the subject population and the period of observation.
In the present study, 50% of APOE e4 carriers with MCI
converted to AD while 40% of APOE e4 non-carriers with MCI
converted. This supports that the progressive value of genetic
assessment alone is comparatively low although a high percentage
of progressive MCI patients carry the APOE e4 allele [18]. In
contrast, this study showed that 83.3% of APOE e4/4 carriers
with MCI converted to AD. These findings indicate that the
presence of the APOE e4/4 in MCI patients is strongly associated
with conversion to AD. The homozygous carriers of the APOE e4
allele could have an increased risk of MCI progression to AD.
The present study demonstrated, using [11C]-PIB PET
imaging, that 56.8% of 51 MCI patients with Ab deposition
converted to AD over 2 years, compared to 5.8% of 17 MCI
patients without Ab deposition. In addition, when MCI patients
with Ab deposition, aged 75289 years, had an impaired episodic
memory, the rate of MCI conversion was 80%. It has previously
been reported that 82% of PIB-positive amnestic subjects with
MCI converted to AD over a 3-year follow-up period [15], and
67% of MCI subjects with high PIB retention converted over 2
years [16]. These findings indicate that a high proportion of MCI
patients with Ab deposition convert to AD within a short period.
In vivo detection of Ab deposition with PET imaging is a useful
prognostic tool for identifying MCI progression to AD.
We found that MCI converters had the regional hypometabolism in cortical regions of lateral temporal cortex, parietal cortex
and precuneus at baseline compared to stable patients. Our results
are consistent with the previous study that regional metabolic
reduction, in temporoparietal or posterior cingulate cortices, was
detected at baseline in MCI patients who converted to AD within
1 year, compared with non-converters [17]. The AD pathophysiological process has been demonstrated to be temporoparietal
and/or precuneus hypometabolism, which is a topographic
pattern characteristic of AD. Therefore, if a regional hypometabolism in the temporal, parietal and/or precuneus cortices is
detected on FDG PET imaging, the progression to AD may occur
within a few years.
An evaluation in predictive accuracy by combining different
biomarkers and neuropsychological variables for the prediction of
AD in MCI has been reported [19]. Inconsistencies in recent
findings are likely due to a variability in neuroimaging processing
techniques, CSF protein immunoassay techniques, setting cutoff
Acknowledgments
GE Healthcare Ltd. (Little Chalfont, Buckinghamshire, HP7 9NA, United
Kingdom) entered into the sublicense agreement with the Shonan-Atsugi
Hospital to allow the use of the [11C]-PIB PET imaging. The authors
thank all the patients and their relatives who have participated in this study,
and the staff at Shonan-Atsugi Hospital.
Conceived and designed the experiments: SH HY. Performed the
experiments: SH HY. Analyzed the data: SH HY. Contributed
reagents/materials/analysis tools: SH HY. Wrote the paper: SH HY.
References
70
1. Petersen RC, Aise PS, Beckett LA, Donohue MC, Gamst AC, et al. (2010)
Alzheimers Disease Neuroimaging Initiative (ADNI). Neurology 74: 201209.
2. Dubois B, Feldman HH, Jacova C, Dekosky ST, Barberger-Gateauet P, et al.
NINCDS-ADRDA criteria. Lancet Neurol 6: 734746.
3. Albert MS, DeKosky ST, Diskson D, Dubois B, Feldman HH, et al. (2011) The
diagnosis of mild cognitive impairment due to Alzheimers disease: Recommen-
dation from the National Institute on Aging-Alzheimers Association workgroups

on diagnostic guidelines for Alzheimers disease. Alzheimers Dement 7: 270
279.
4. Forsberg A, Engler H, Almkvist O, Blomquist G, Hagman G, et al. (2008) PET

5. Jack CR, Lowe VJ, Weigand SD, Wiste HJ, Senjem ML, et al. (2009) Serial PIB
and MCI in normal, mild cognitive impairment and Alzheimers disease:
implications for sequence of pathologic events in Alzheimers disease. Brain 132:
13551365.
6. Landau SM, Harvey D, Madison CM, Reiman EM, Foster NL, et al. (2010)
Comparing predictors of conversion and decline in mild cognitive impairment.
7. Mathis CA, Wang Y, Holt DP, Huang GF, Debnath ML, et al. (2003) Synthesis
and evaluation of [11C]-labeled 6-substituted 2-arylbenzothiazoles as amyloid
imaging agents. J Med Chem 46: 27402754.
8. Klunk WE, Engler H, Nordberg A, Wang Y, Blomqvist G, et al. (2004) Imaging
brain amyloid in Alzheimer disease with Pittsburgh Compound-B. Ann Neurol
55: 306319.
9. Hatashita S, Yamasaki H (2010) Clinically different stages of Alzheimers disease
associated by amyloid deposition with [11C]-PIB PET imaging. J Alzheimers
Dis 21: 9951003.
10. Folstein MF, Folstein SE, McHugh PR (1975) Mini-mental state A practical
Res 12: 189198.
11. Morris JC (1993) The clinical dementia rating (CDR): current version and
12. Wechsler DA (1987) Wechsler Memory Scale-Revised. New York: Psychological
Corporation. New York.
13. Mckhann G, Drachman DA, Folstein M, KatzmanR, Price D, et al. (1984)
Clinical diagnosis of Alzheimers disease-report of the NINCDS-ADRDA work
14.
15.
16.
17.
18.
19.
20.
group under the auspices of Department of Health and Human Services Task
Force on Alzheimers disease. Neurology 34: 939944.
Lopresti BJ, Klunk WE, Mathis CA, Hoge JA, Ziolko SK, et al. (2005) Simplified
quantification of Pittsburgh Compound B amyloid imaging PET studies: a
comparative analysis. J Nucl Med 46: 19591972.
Okello A, Koivumen J, Edison P, Acher HA, Turkheimer FE, et al. (2009)
Conversion of amyloid positive and negative MCI to AD over 3 years: An
[11C]-PIB PET study. Neurology 73: 754760.
Villemagne VL, Pike KE, Chetelat G, Ellis KA, Mulligan RS, et al. (2011)
Longitudinal Assessment of AB and cognition in Aging and Alzheimer Disease.
Ann Neurol 69: 181192.
Drzezga A, Lautenschlager N, Siebner H, Riemenschneider M, WiIloch F, et al.
(2003) Cerebral metabolic change accompanying conversion of mild cognitive
impairment into Alzheimers disease: a PET follow-up study. Eur J Nucl Med
Mol Imaging 30: 11041113.
Drzezga A, Grimer T, Riemenschneider M, Lautenschlager N, Siebner H, et al.
(2005) Prediction of individual clinical outcome in MCI by means of genetic
assessment and [18F]-FDG PET. J Nucl Med 46: 16251632.
Ewers M, Walsh C, Trojanowski JQ, Shaw LM, Peterson RC, et al. (2012)
Prediction of conversion from mil cognitive impairment to Alzheimers disease
dementia based upon biomarkers and neuropsychological test performance.
Zhang S, Han D, Tan X, Feng J, Guo Y, et al. (2012) Diagostic accuracy of
[18F]-FDG and [11C]-PIB-PET for prediction of short-term conversion to
Alzheimers disease in subjects with mild cognitive impairment. Int J Clin Pract
66: 185198.
71
Regional Amyloid Deposition in Amnestic Mild Cognitive

Impairment and Alzheimers Disease Evaluated by
[18F]AV-45 Positron Emission Tomography in Chinese
Population
Kuo-Lun Huang1., Kun-Ju Lin2,3., Ing-Tsung Hsiao2,3, Hung-Chou Kuo1, Wen-Chuin Hsu1, WenLi Chuang1, Mei-Ping Kung2,4, Shiaw-Pyng Wey2,3, Chia-Ju Hsieh2,3, Yau-Yau Wai5, Tzu-Chen Yen2*, ChinChang Huang1*
1 Department of Neurology and Dementia Center, Chang Gung Memorial Hospital and Chang Gung University, Taoyuan, Taiwan, 2 Molecular Imaging Center and
Department of Nuclear Medicine, Chang Gung University and Chang Gung Memorial Hospital, Taoyuan, Taiwan, 3 Healthy Aging Research Center and Department of
Medical Imaging and Radiological Sciences, College of Medicine, Chang Gung University, Taoyuan, Taiwan, 4 Department of Radiology, University of Pennsylvania,
Pennsylvania, United States of America, 5 Department of Radiology, Chang Gung Memorial Hospital, Taoyuan, Taiwan
Abstract
Background: To compare the neocortical amyloid loads among cognitively normal (CN), amnestic mild cognitive
impairment (aMCI), and Alzheimers disease (AD) subjects with [18F]AV-45 positron emission tomography (PET).
Materials and Methods: [18F]AV-45 PET was performed in 11 CN, 13 aMCI, and 12 AD subjects to compare the cerebral
cortex-to-whole cerebellum standard uptake value ratios (SUVRs) of global and individual volumes of interest (VOIs) cerebral
cortex. The correlation between global cortical [18F]AV-45 SUVRs and Mini-Mental State Examination (MMSE) scores was
analyzed.
Results: The global cortical [18F]AV-45 SUVRs were significantly different among the CN (1.0860.08), aMCI (1.2760.06), and
AD groups (1.3460.13) (p = 0.0003) with amyloidosis positivity rates of 9%, 62%, and 92% in the three groups respectively.
Compared to CN subjects, AD subjects had higher SUVRs in the global cortical, precuneus, frontal, parietal, occipital,
temporal, and posterior cingulate areas; while aMCI subjects had higher values in the global cortical, precuneus, frontal,
occipital and posterior cingulate areas. There were negative correlations of MMSE scores with SUVRs in the global cortical,
precuneus, frontal, parietal, occipital, temporal, posterior cingulate and anterior cingulate areas on a combined subject pool
of the three groups after age and education attainment adjustment.
Conclusions: Amyloid deposition occurs relatively early in precuneus, frontal and posterior cingulate in aMCI subjects.
Higher [18F]AV-45 accumulation is present in parietal, occipital and temporal gyri in AD subjects compared to the aMCI
group. Significant correlation between MMSE scores and [18F]AV-45 SUVRs can be observed among CN, aMCI and AD
subjects.
Citation: Huang K-L, Lin K-J, Hsiao I-T, Kuo H-C, Hsu W-C, et al. (2013) Regional Amyloid Deposition in Amnestic Mild Cognitive Impairment and Alzheimers
Disease Evaluated by [18F]AV-45 Positron Emission Tomography in Chinese Population. PLoS ONE 8(3): e58974. doi:10.1371/journal.pone.0058974
Editor: Karl Herholz, University of Manchester, United Kingdom
Received November 28, 2012; Accepted February 8, 2013; Published March 14, 2013
Copyright: 2013 Huang et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
Funding: This study was carried out with financial support from the National Science Council, Taiwan (Grant NSC-97-2314-B-182A-050, NSC-98-2314-B-182-056,
NSC-100-2314-B-182-003, NSC-100-2314-B-182A-091-MY3, and NSC-101-2314-B-182A-061-MY2), and Chang Gung Memorial Hospital (Grant CMRPG300071,
CMRPG390793, and CMRPD1A0312). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
* E-mail: [email protected] (CCH); [email protected] (TCY)
Introduction
resulting in excessive fibrillary amyloid plaque accumulation in the

brain. Furthermore, the accumulation of amyloid plaques may
precede the diagnosis of AD by 10 years [2]. Therefore, a reliable
biomarker for determination of amyloid plaque aggregates could
facilitate an early diagnosis and treatment of AD.
Molecular neuroimaging modalities such as positron emission
tomography (PET) have a potential to provide functional
information on the pathological process of AD. Several PET
ligands have been investigated to determine their ability for

and leads to progressive dysfunctions in memory and other
cognitive domains. The diagnosis of AD is hampered by a lack of
noninvasive biomarkers. Moreover, 67% of dementia patients with
mild clinical symptoms may not be diagnosed by physicians [1].
The most widely accepted pathogeneses of AD include increased
production or impaired clearance of amyloid-beta oligomers,
72
March 2013 | Volume 8 | Issue 3 | e58974

Amyloid Distribution Patterns on [18F]AV-45 PET
detecting amyloid plaques, including [11C]PIB [3], [11C]SBI-13

[4], [11C]BF-227 [5], [11C]MeS-IMPY [6], [18F]FDDNP [7],
[18F]AV-45 (florbetapir) [8], [18F]AV-1 (florbetaben, BAY949172) [9], and [18F]GE067 (flutemetamol) [10]. At present, the
most widely researched imaging technique utilizes [11C]PIB to
detect amyloid plaques, but the short half-life of the carbon-11
(approximately 20.4 min) of [11C]PIB limits its clinical application.
On the other hand, PET tracers utilizing fluorine-18 have been
designed to overcome this limitation since this isotope has a longer
half-life of about 109.4 min, which makes off-site preparation and
regional distribution easier. Among these tracers, [18F]AV-45 has
gained great attention as a PET imaging agent targeting amyloid
plaques and been selected as the standard plaque imaging agent
for the Alzheimers Disease Neuroimaging Initiative (ADNI) [11].
[18F]AV-45 shows good amyloid-binding affinity with a desirable
pharmacokinetics. It remains at steady levels in the brain 30
90 min after injection, and a short imaging time (510 min) is
sufficient [8]. A recent phase III study with 35 terminally ill
participants was reported to compare [18F]AV-45 imaging and
postmortem amyloid immunohistochemistry. The results showed
that [18F]AV-45 imaging is well correlated with postmortem
amyloid immunohistochemistry [12].
In this preliminary study, we aimed to compare the amyloid
plaque distribution on [18F]AV-45 PET scan in cognitively normal
(CN), amnestic mild cognitive impairment (aMCI) and AD
subjects, and correlate the cerebral cortical amyloid load with
cognitive functioning among CN, aMCI and AD subjects. In
addition, we aimed to determine a cutoff value to differentiate CN
subjects from AD subjects in a Chinese population.
their next of kin, caregivers or guardians if subjects were clinically

suspected of AD. The written informed consent was obtained from
all subjects before their participation in the study. In addition, the
next of kin or guardians of 6 AD patients also gave their written
informed consents since these patients capacity to comprehend
the study protocol was compromised or they could not sign their
names clearly. The study protocol and the procedure of obtaining
informed consents were approved by the institutional review board
of Chang Gung Memorial Hospital and the Governmental
Department of Health.
Magnetic Resonance Imaging (MRI) Acquisition

The MRI images were acquired on a 3T MR scanner
(Magnetom Trio, a TIM system; Siemens, Erlangen, Germany).
The scanning protocol included an axial fluid attenuation
inversion
recovery
(FLAIR)
sequence
(TR = 9000 ms,
TE = 87 ms, TI = 2500 ms, voxel size = 0.960.764 mm3) and a
whole brain axial three-dimensional T1-weighted magnetization
prepared rapid acquisition gradient echo (MP-RAGE) sequence
(TR = 2000 ms, TE = 2.63 ms, TI = 900 ms, flip angle = 9u, voxel
size = 16161 mm3), which was subsequently reformatted as
planes perpendicular to the long axis of the hippocampus in
2 mm slice thickness. An additional coronal T2-weighted turbo
spin echo sequence (TR = 7400 ms, TE = 95 ms, voxel size = 0.460.462 mm3) was acquired with the identical geometric
orientation with the reformatted coronal T1-weighted images [8].
[18F]AV-45 PET Imaging

[18F]AV-45 was synthesized at the cyclotron facility of Chang
Gung Memorial Hospital following the method described previously [18]. In brief, [18F]fluoride was obtained by bombardment
of 18O-enriched water with 12-MeV protons accelerated in a SHI12S cyclotron (Sumitomo Heavy Industries, Tokyo, Japan). The
[18F]fluoride was then trapped on an Accel Plus QMA cartridge
(Waters (Milford, MA) and eluted to a reaction vessel using a
mixture of K2CO3, Kryptofix2.2.2 and acetonitrile. Following
azeotropic distillation, the dehydrated [18F]fluoride was reacted
with AV-105, the precursor for synthesis of [18F]AV-45 provided
by Avid Radiopharmaceiticals (Philadelphia, PA), to form a
fluorinated intermediate, N-Boc-[18F]AV- 45, which was in turn to
be hydrolyzed using HCl to remove the protection group. Product
[18F]AV- 45 was obtained after purification with semi-preparative
HPLC. The radiochemical purity of [18F]AV-45 was greater than
95%, and the specific activity was greater than 4000 TBq/mmol at
the end of synthesis (EOS). The mean administered activity was
365 MBq (range, 325394 MBq). The PET acquisition protocol
and optimal scanning time were adapted from previous reports
[8,19]. In brief, helical computed tomography images were
obtained for attenuation correction at 40 min. Each PET
acquisition consisted of two 5-min dynamic frames obtained 50
60 min post-injection in 3D mode [19] from the Biograph mCT
PET/CT System (Siemens Medical Solutions, Malvern, PA,
USA). PET projection data were iteratively reconstructed using
3-D OSEM algorithm of 4 iterations, 24 subsets, postsmoothed by
a Gaussian filter of 2 mm FWHM and zoom = 3, and with CTbased attenuation correction. Scatter and random correction were
also performed using the correction methods provided by the
manufacture. The reconstructed images were with pixel size of
0.679 mm, and slice thickness of 2.027 mm. Individual frames of
the [18F]AV-45 PET dynamic series were realigned if motion was
detected. Summed images were subsequently created for analysis
[20].
Methods
Participants
Thirty-six subjects were studied at the Center for Dementia of
the Linkou Chang Gung Memorial Hospital (CGMH), Taiwan,
including 11 CN, 13 aMCI, and 12 AD subjects. All subjects
underwent a neurological examination, neurocognitive evaluation,
and routine blood analysis. Study subjects were categorized on the
basis of the consensus of panels composed of neurologists,
neuropsychologists, neuroradiologists and experts in nuclear
medicine. The CN subjects were volunteers with normal cognitive
performance; the methods for CN subject recruitment have been
described previously [8]. The diagnostic criteria for aMCI were
based on those proposed by Petersen et al [13]: (1) subjective
memory complaints by the patient or an informant, (2) relatively
normal performance in other cognitive domains and (3) normal
activities of daily living, (4) objective memory impairment on at
least one neurocognitive test of memory performance, and (5) no
dementia according to DSM-IV criteria [14]. No rigid cutoff score
was applied to determine objective memory impairment, but
aMCI was generally determined when memory measures fell 1.0
1.5 standard deviations below the means for age-matched norms
in Taiwan. The diagnosis of AD was made when subjects fulfilled
the National Institute of Neurological and Communicative
Disorders and StrokeAlzheimers Disease and Related Disorders
Association (NINCDS-ADRDA) criteria for probable AD [15]. A
comprehensive battery of neurocognitive tests was administered to
obtain objective evidence of cognitive impairment, including the
Mini-Mental State Examination (MMSE) [16], Clinical Dementia
Rating (CDR), logical memory in the Wechsler memory scalerevised (LM), visual-association memory test (VAMT), category
verbal fluency test (CVFT), trail-making A test (TMAT), and
clock-drawing test (CDT) [17]. Prior to enrollment, the study
objectives and protocol were well informed to all subjects and also
73
2
[18F]AV-45 Imaging Processing and Atlas-based

Quantitative Volume of Interest (VOI) Analysis
multiple comparison test. Correlations between cognitive test

scores and cortical SUVRs were analyzed with both Pearson
correlations and partial regression with adjustment for age and
years of education. Statistical analyses were performed with SAS
version 9.2 (SAS Institute Inc., New York, USA), and p,0.05 was
considered significant.
Both PET and MRI images from each subject were coregistered first, and then MRI images were spatially normalized
into the MNI T1 template as in Statistical Parametric Mapping 5
(SPM5) (Wellcome Department of Cognitive Neurology, University College London, London, UK). The transformation parameters of MR normalization were then applied to corresponding
PET images. All the image data were processed in the PMOD
software (version 3.2; PMOD Technologies Ltd., Zurich, Switzerland). For all subjects, a binary gray matter mask was first created
by taking the non-zero pixels from an averaged gray matter
probability map generated from the segmentation of the spatially
normalized MR images from normal subjects. The final VOI
template for VOI analysis was then obtained by taking the
intersection of both the binary gray matter mask and the
automated anatomic labeling (AAL) atlas (VOI template) for all
subjects [21,22]; this step was performed to minimize the inclusion
of both cerebrospinal fluid and white matter (and thus non-specific
white matter [18F]AV-45 retention) in the statistical measures of all
VOIs [23]. The original AAL atlas which contains 116 regions was
combined into the following 8 labeled VOIs for further analysis:
the precuneus, frontal, parietal, occipital, temporal, posterior
cingulate and anterior cingulate areas, and the whole cerebellum.
The standard uptake value ratio (SUVR) was calculated by
determining the ratios of the integrated activities between the
target VOI (cortical regions) and whole cerebellum reference VOI.
Voxel-based statistical analysis was conducted with SPM5 [24].
The spatial normalization of individual [18F]AV-45 images to
MNI MRI T1-template space was performed in the same way as
that for the VOI analysis described above. All voxels in the
normalized [18F]AV-45 images were divided by the mean
[18F]AV-45 uptake of the whole cerebellum VOI in each subject
to form SUVR images. Moreover, all PET images in SPM analysis
were smoothed by 8 mm FWHM Gaussian filtering. Voxel-wise
[18F]AV-45 uptake differences among the CN, aMCI, and AD
subjects were assessed in SPM5. Statistical maps displaying group
differences were shown at a significance value of p,0.01 and
extent voxels of 100. Due to a relatively small sample size of our
study, and to increase the sensitivity in the SPM analysis, the lower
threshold of two-tailed p,0.01 under uncorrected statistics was
applied here [25].
Results
The study recruited 11, 13, and 12 subjects in the CN, aMCI,
and AD groups, respectively. The demographic data were shown
in Table 1. None of the subjects had adverse events directly related
to [18F]AV-45 administration.
VOI Analysis
The VOI analysis of [18F]AV-45 images revealed significant
differences in the global cortical SUVR among subjects with CN,
aMCI and AD (Figure 1). Regarding to the differences in regional
SUVRs, the precuneus, frontal, parietal, occipital, temporal, and
posterior and anterior cingulate cortical SUVRs were also
different among the three groups. In the post hoc analyses, all of
these areas except the anterior cingulate had a higher SUVR of
[18F]AV-45 in AD subjects than CN subjects. Furthermore, the
SUVRs of the global cortical, precuneus, frontal, occipital and
posterior cingulate areas were higher in aMCI subjects than CN
subjects. There was no difference in [18F]AV-45 SUVRs of these
predefined areas between aMCI and AD subjects. With the SUVR
cutoff point of 1.178, the positive rate of cerebral cortical
amyloidosis was 8%, 62%, and 92% for CN, aMCI and AD
groups, respectively. Interestingly, there were varying degrees of
[18F]AV-45 retention in CN, aMCI negative for cerebral
amyloidosis (aMCI (2)), aMCI positive for cerebral amyloidosis
(aMCI (+)), and AD subjects, whose patterns closely followed the
amyloid deposition distribution observed in post mortem brain
tissue (Figure 2) [26].
SPM Analysis
The SPM analysis showed that aMCI subjects had significantly
higher [18F]AV-45 uptake than CN subjects in the frontal,
temporal, parietal, occipital areas, as well as in the precuneus,
which showed the most prominent differences (Figure 3A). These
results were supported by the automated VOI analysis in which
aMCI patients showed 15% increased [18F]AV-45 uptake in the
Determination of the Threshold for Cerebral Cortical

Amyloidosis in [18F]AV-45 Images
Table 1. Demographic characteristics of the CN, aMCI, and

AD groups.
To quantitatively measure the threshold for global cerebral

cortical amyloid burden, the [18F]AV-45 images of the AD and
CN groups were compared to derive the cutoff point with receiver
operating characteristic (ROC) analysis. Subjects with a cortical
SUVR level exceeding the threshold were categorized as positive
for cerebral amyloidosis. In our cases, the threshold derived from
the AD and CN groups was 1.178 with the area under the curve of
0.96 (p,0.001); the sensitivity and specificity were 92% and 91%,
respectively.
Statistical Analyses
74
Data were expressed as means6SD or absolute numbers with

proportions for descriptive statistics. Continuous variables were
analyzed by non-parametric statistics with the MannWhitney Utest. Categorical data were analyzed using Fishers exact test.
Differences of [18F]AV-45 cortical SUVRs between groups were
assessed by non-parametric one-way analysis of variance (ANOVA) using the KruskalWallis test followed by Dunns post hoc
CN (n = 11)
aMCI (n = 13)
AD (n = 12)
Gender (M/F)
6/5
5/8
3/9
0.35
Age, years
69.3 (6.6)
70.0 (11.2)
75.8 (5.7)
0.15
CDR
0.0 (0.0)
0.5 (0.0)*
1.1 (0.3)*
,0.0001
MMSE
27.9 (1.6)
19.7 (2.4)*
12.9 (3.7)*
,0.0001
Memory deficit duration, years
3.0 (1.2)
5.9 (2.2)
,0.001
Education level, 9.4 (4.1)

years
6.2 (4.1)
7.2 (4.3)
0.19
*p,0.05 compared to the CN group.

Values were mean (SD) or numbers of patients.
CN, cognitively normal; aMCI, amnestic mild cognitive impairment; AD,
Alzheimers disease; CDR, clinical dementia rating; MMSE, mini-mental state
examination.

Figure 1. Scatter plots showing general and individual region-to-cerebellum ratios between different groups. Comparisons of [18F]AV45 SUVR in the predefined VOIs of the global cortical (A), precuneus (B), frontal (C), parietal (D), occipital (E), temporal (F), posterior cingulate (G) and
anterior cingulate (H) areas among the cognitively normal (CN), amnestic mild cognitive impairment (aMCI), and Alzheimers disease (AD) subjects. (*
p,0.05, ** p,0.005, *** p,0.0005, when comparing SUVR among CN, aMCI and AD subjects with the KruskalWallis ANOVA test. { p,0.05 versus CN
subjects with Dunns post hoc analysis).
Correlations between [18F]AV-45 SUVRs and Cognitive

Function
frontal cortex (p,0.05) and 30% in the precuneus (p,0.005) than

the CN group. Individually, in the precuneus, 8 of 13 aMCI
subjects had [18F]AV-45 uptake values above 2 SD from the CN
group mean. Comparing the SUVRs between AD and aMCI
subjects, AD subjects had a slightly increased [18F]AV-45 uptake
over the frontal, temporal and parietal regions (Figures 3B and
3C). Correspondingly, the VOI analysis showed significant
increase of [18F]AV-45 in AD as compared to CN subjects in
the additional brain regions including parietal (30% increase from
the CN group mean, p,0.01), occipital (26% increase from the
CN group mean, p,0.01), and temporal regions (20% increase
from the CN group mean, p,0.01). Tables S1, S2 and S3 listed
the peaks of the most significant increase in [18F]AV-45 uptake
obtained in this analysis with Talairach and Tournoux coordinates, x, y, and z in millimeters of these peaks, as well as the
corresponding z scores.
There was no association of the global cortical [18F]AV-45

SUVR with age and education attainment. There was a negative
correlation of the global cortical [18F]AV-45 SUVR with the
scores of MMSE (adjusted R2 = 0.44), LM (adjusted R2 = 0.44),
CVFT (adjusted R2 = 0.37), TMAT (adjusted R2 = 0.48), CDT
(adjusted R2 = 0.56) and VAMT (adjusted R2 = 0.57) with all p
values ,0.001 after adjustment for age and education. Furthermore, there was a negative correlation of MMSE scores with all
the predefined VOI areas after adjustment for age and education
(Figure 4).
Figure 2. Representative sagittal [18F]AV-45 PET images. Varying degrees of [18F]AV-45 PET uptake in CN, aMCI negative for cerebral
amyloidosis (aMCI (-)), aMCI positive for cerebral amyloidosis (aMCI (+)), and AD subjects (bottom row) were closely corresponding to the pathological
amyloid deposition distribution adapted from Braak and Braak (top row) [26].
75
Figure 3. Statistical parametric mapping analysis: Localization of increased [18F]AV-45 retention between CN, aMCI and AD
subjects. Comparisons of [18F]AV-45 SUVRs between cognitively normal (CN) and amnestic mild cognitively impairment (aMCI) subjects (A), and
between aMCI and Alzheimers disease (AD) subjects (B) (Uncorrected for multiple comparisons and the color bar values indicate the value of the Tstatistic in each display). Surface rendering was used to illustrate the cortical areas where [18F]AV-45 SUVRs were increased in aMCI than CN subjects
(red) and increased in AD than aMCI subjects (green) (C).
fluorine-18-labeled radiotracers, including [18F]AV-45 (florbetapir), [18F]AV-1 (florbetaben) and [18F]GE067 (flutemetamol), have
been extensively employed for amyloid plaque detection both in
vivo and ex vivo [29]. Although [11C] PIB PET has been applied
to compare the amyloid plaque distribution among CN, MCI and
AD subjects in the Asian population [30], there are limited reports
about fluorine-18-labeled amyloid PET studies in non-Caucasians
[20,31,32]. In a recent report by Barthel et al, the Caucasian
elderly normal subjects had higher SUVRs in different neocortical
Discussion
Although the cause of AD is still unknown, the presence of
amyloid plaques plays a pivotal role in the pathogenesis of the
disease. Studies show that the presence of amyloid plaques in
patients with mild cognitive impairment may increase the potential
of conversion to AD [27]. Therefore, the development of imaging
techniques to detect amyloid plaques in vivo would be helpful for
early diagnosis and monitoring disease progress [28]. Several
76
Figure 4. Relationship between MMSE scores and global and regional cortical [18F]AV-45 retention. Results of the negative regressions
between MMSE scores and global and regional cortical [18F]AV-45 SUVRs among the cognitive normal, amnestic mild cognitive impairment, and
Alzheimers disease subjects with age and years of education as covariates.

regions as compared to the Japanese elderly normal subjects on

the florbetaben PET [31]. To our knowledge, this is the first
[18F]AV-45 PET study to compare the amyloid loads among CN,
aMCI, and AD subjects in vivo in an Eastern population. We
found that the [18F]AV-45 SUVR was markedly increased in
aMCI and AD subjects compared to CN subjects, regardless of
whether it was measured with the VOI or SPM methods. In
addition, we also applied the ROC method to determine the
threshold for global cortical amyloid positivity, which was 1.178
with a sensitivity of 92% and a specificity of 91%, respectively.
Our findings are concordant with those of Fleisher et al [33]. They
applied the data from another post-mortem study to determine the
[18F]AV-45 threshold for amyloid positivity, and the pathologybased threshold was 1.17, which is similar to our clinical diagnosisbased threshold of 1.178.
At present, the diagnosis of AD is mainly based on clinical
manifestations, brain structure imaging, laboratory study, and
neuropsychological assessment. Around 1020% of clinically
diagnosed AD patients do not have amyloid plaques on postmortem pathological examination [34,35]. In our study, the
positivity of amyloidosis in the CN, aMCI, and AD groups was
9%, 62%, and 92%, respectively. The positivity for amyloidosis in
AD patients (92%) is relatively high compared to the 80.9%
observed by Fleisher et al [33]. One of the reasons is that our AD
subjects had a relatively low cognitive performance and education
attainment. Besides, the average disease duration was 5.962.2
years. Interestingly, 1 of the 12 AD patients (8%) had a relatively
low global cortical SUVR, with a level similar to that of the CN
subjects in our study, despite the fact that the patient exhibited a
typical presentation of AD and fulfilled the most widely applied
NINCDS-ADRDA criteria. However, the criteria may still allow a
degree of uncertainty given the fact of the lack of pathological
findings. On the other hand, PET radiotracers may fail to bind
amyloid plaques because different fibrillary amyloid aggregate
forms may have different affinities for amyloid radiotracers
[36,37].
The accumulation of amyloid plaques is a relentless process with
a stereotypic progression pattern in the disease course of AD. In
our study, the distribution of [18F]AV-45 uptake in the aMCI
group was intermediate between that of the CN and AD groups.
Comparing the regional [18F]AV-45 SUVRs between aMCI and
CN subjects, a higher [18F]AV-45 SUVR was noted over the
precuneus, frontal, occipital and posterior cingulate areas. These
findings were similar with the report from Kempainen et al that
aMCI patients had higher [11C]PIB retention in the frontal,
parietal, temporal, and posterior cingulate areas under SPM
analysis [38]. Furthermore, the averaged [18F]AV-45 SUVR
distribution pattern in aMCI subjects positive for amyloidosis was
more like Stage B by the Braak and Braaks amyloid deposition
category, while the distribution pattern in aMCI subjects negative
for amyloidosis was more like Stage A (Figure 2) [26].
Additionally, a more widely spread [18F]AV-45 uptake was noted
over the frontal, temporal, and parietal areas in AD subjects than
aMCI subjects in SPM analysis, which are concordant with those
of the previous [18F]AV-45 and [11C]PIB studies [19,39,40].
Although the [18F]AV-45 distribution among the CN, aMCI and
AD groups could mirror the pathological staging, its association
with cognitive decline needs to be followed up in further cohort
studies.
The deposition of amyloid plaques has been reported to occur
at least 10 years prior to the clinical diagnosis of AD; about 21% of
non-demented elderly have amyloid pathology [41]. However, the
positivity of cerebral amyloidosis was relatively low in our CN
group. Age is associated with amyloid load, and the relatively
young age in our CN group may contribute to the low positivity

[42].
Mild cognitive impairment is a transitional state between
normal cognition and AD; it can be present years before the
diagnosis of AD. The amyloid load in subjects with MCI is
correlated with the rate of conversion from MCI to AD, suggesting
that it is a prognostic marker in MCI subjects [27]. However,
many clinical and genetic factors, such as age, APOE4 allele
carrier, and cerebral atrophy severity, may also influence the
amyloid load in MCI subjects [17,42,43]. Therefore, the
distribution of amyloid load in MCI subjects could be diverse,
and a great overlap with CN and AD groups can be expected
[33,44]. In our study, AD subjects tended to have a higher global
cortical [18F]AV-45 SUVR (1.3460.13) than aMCI patients
(1.2760.16) with an effect size of 0.45, but the sample size in this
study is not large enough to differentiate the 2 groups based on the
VOI analysis. Nonetheless, a more extensive [18F]AV-45 retention
could be noted over the frontal, parietal, temporal and occipital
areas under the voxelwise SPM analysis. On the other hand,
amyloid deposition is an early change in the pathogenesis of AD.
Rather than the cortical amyloid load, the downstream pathway
such as synaptic dysfunction and brain structure change may play
a more important role in determining the cognitive declines from
MCI to AD [2,27], and the application of multimodal neurochemical and imaging biomarkers should be considered to early
identify predementia and even preclinical asymptomatic stages of
AD [45].
Alzheimers disease is characteristic of progressive amyloid
plaques accumulation accompanied with cognitive decline. Roe et
al found that mean cortical [11C]PIB binding potential was not
only related to MMSE scores, but also the cognitive measures
associated with executive functions among AD and CN subjects
[46]. However, they didnt correlate the regional cortical [11C]PIB
binding potential with the various cognitive measures. In the study
by Barthel et al, both the frontal and total brain amyloid volumes
on [18F]AV-1 PET were negatively correlated with MMSE scores
among AD and CN subjects [47]. These findings are consistent
with our study results that there was a negative correlation
between MMSE scores and [18F]AV-45 SUVRs of various cortical
regions, especially over the frontal, temporal and precuneus areas
even after controlling for age and years of education. The presence
of negative correlation between MMSE scores and the SUVRs of
different amyloid radiotracers on PET may support the cascade
hypothesis that amyloid plaque accumulation plays a pivotal role
in the AD pathophysiological process [48].
There were several limitations in the current study. Correcting
partial volume effect (PVE) is important for VOI-based analysis
and in particular, for amyloid imaging. Therefore, the first
limitation in this study is that no PVE correction was performed.
The use of binary gray matter mask in this work is to avoid
extracting activity from both CSF and white matter regions by
confining VOI quantification only to gray matter region. To
improve quantification accuracy in small VOIs, PVE correction
should be included as a future work. Second, the sample size in our
study is relatively small, and the results are not generalizable to the
Chinese population. In addition, a cutoff value of 1.178 for
neocortical [18F]AV-45 uptake was derived to differentiate
Chinese AD from CN subjects, and its application to predict the
conversion to AD from CN or aMCI subjects has to be verified in
a longitudinal cohort study.
77
6
of the most significant increased [18F]AV-45 uptake in aMCI

patients than CN subjects, p,0.01 (unc.), extent voxels = 100.
(DOC)
Conclusions
The findings from our study show that [18F]AV-45 PET can be
applied in vivo to differentiate the cerebral amyloid burden among
CN, aMCI, and AD subjects using continuous or binary
measurements, and [18F]AV-45 SUVR is associated with MMSE
scores. The clinical diagnosis-based threshold for cerebral
amyloidosis derived from AD patients is similar to the pathology-based threshold. Therefore, [18F]AV-45 PET is a potential
biomarker for AD in multicenter monitoring and treatment trials.
Table S3 Comparing [18F]AV-45 uptake between Alz-
heimers disease (AD) patients and cognitively normal

(CN) subjects. The locations and values of the most significant
increased [18F]AV-45 uptake in AD patients than CN subjects,
p,0.01 (unc.), extent voxels = 100.
(DOC)
Acknowledgments
We would like to thank Ko-Ting Chao and Yu-Chen Hsieh for their
technical assistance. We also thank Avid Radiopharmaceuticals, Inc.
(Philadelphia, PA, USA) for providing the precursor for the preparation of
[18F]AV-45.
Table S1 Comparing [18F]AV-45 uptake between Alz-
heimers disease (AD) patients and amnestic mild

cognitive impairment (aMCI) patients. The locations and
values of the most significant increased [18F]AV-45 uptake in AD
patients than aMCI patients, p,0.01 (unc.), extent voxels = 100.
(DOC)
Conceived and designed the experiments: KLH MPK TCY CCH.
Performed the experiments: KLH KJL HCK WCH WLC TCY CCH.
Analyzed the data: KLH KJL ITH MPK SPW CJH YYW TCY CCH.
Contributed reagents/materials/analysis tools: KLH KJH ITH TCY
CCH. Wrote the paper: KLH KJL MPK YYW TCY CCH.
Comparing [18F]AV-45 uptake between amnestic mild cognitive impairment (aMCI) patients and
cognitively normal (CN) subjects. The locations and values
Table S2
References
78
1. Lopponen M, Raiha I, Isoaho R, Vahlberg T, Kivela SL (2003) Diagnosing

cognitive impairment and dementia in primary health care a more active
approach is needed. Age Ageing 32: 606612.
2. Sperling RA, Aisen PS, Beckett LA, Bennett DA, Craft S, et al. (2011) Toward
defining the preclinical stages of Alzheimers disease: recommendations from the
National Institute on Aging-Alzheimers Association workgroups on diagnostic
guidelines for Alzheimers disease. Alzheimers Dement 7: 280292.
3. Klunk WE, Wang Y, Huang GF, Debnath ML, Holt DP, et al. (2001)
Uncharged thioflavin-T derivatives bind to amyloid-beta protein with high
affinity and readily enter the brain. Life Sci 69: 14711484.
4. Ono M, Cheng Y, Kimura H, Cui M, Kagawa S, et al. (2011) Novel 18F-labeled
benzofuran derivatives with improved properties for positron emission
tomography (PET) imaging of b-Amyloid plaques in Alzheimers brains. J Med
Chem 54: 29712979.
5. Kudo Y, Okamura N, Furumoto S, Tashiro M, Furukawa K, et al. (2007) 2-(2[2-dimethylaminothiazol-5-yl]ethenyl)-6- (2-[fluoro]ethoxy) benzoxazole: a Novel PET agent for in vivo detection of dense amyloid plaques in Alzheimers
disease patients. J Nucl Med 48: 553561.
6. Seneca N, Cai L, Liow JS, Zoghbi SS, Gladding RL, et al. (2007) Brain and
whole-body imaging in nonhuman primates with [11C]MeS-IMPY, a candidate
radioligand for b-amyloid plaques. Nucl Med Biol 34: 681689.
7. Shoghi-Jadid K, Barrio JR, Kepe V, Wu HM, Small GW, et al. (2005) Imaging
b-amyloid fibrils in Alzheimers disease: a critical analysis through simulation of
amyloid fibril polymerization. Nucl Med Biol 32: 337351.
8. Lin KJ, Hsu WC, Hsiao IT, Wey SP, Jin LW, et al. (2010) Whole-body
biodistribution and brain PET imaging with [18F]AV-45, a novel amyloid
imaging agent - a pilot study. Nucl Med Biol 37: 497508.
9. Rowe CC, Ackerman U, Browne W, Mulligan R, Pike KL, et al. (2008) Imaging
of amyloid beta in Alzheimers disease with 18F-BAY94-9172, a novel PET
tracer: proof of mechanism. Lancet Neurol 7: 129135.
10. Koole M, Lewis DM, Buckley C, Nelissen N, Vandenbulcke M, et al. (2009)
Whole-body biodistribution and radiation dosimetry of 18F-GE067: a
radioligand for in vivo brain amyloid imaging. J Nucl Med 50: 818822.
11. Jagust WJ, Bandy D, Chen K, Foster NL, Landau SM, et al. (2010) The
Alzheimers Disease Neuroimaging Initiative positron emission tomography
core. Alzheimers Dement 6: 221229.
12. Clark CM, Schneider JA, Bedell BJ, Beach TG, Bilker WB, et al. (2011) Use of
Florbetapir-PET for Imaging b-Amyloid Pathology. JAMA 305: 275283.
13. Petersen RC, Doody R, Kurz A, Mohs RC, Morris JC, et al. (2001) Current
concepts in mild cognitive impairment. Arch Neurol 58: 19851992.
14. American Psychiatric Association (1994) Diagnostic and statistical manual of
mental disorders, DSM-IV. Washington, DC: American Psychiatric Association.
Clinical diagnosis of Alzheimers disease. Neurology 34: 939.
16. Folstein MF, Folstein SE, McHugh PR (1975) Mini-mental state: a practical
method for grading the cognitive state of patients for the clinician. J Psychiat Res
12: 189198.
17. Wang PN, Hong CJ, Lin KN, Liu HC, Chen WT (2011) APOE M4 increases the
risk of progression from amnestic mild cognitive impairment to Alzheimers
disease among ethnic Chinese in Taiwan. J Neurol Neurosurg Psychiatry 82:
165169.
18. Yao CH, Lin KJ, Weng CC, Hsiao IT, Ting YS, et al. (2010) GMP-compliant
automated synthesis of [(18)F]AV-45 (Florbetapir F 18) for imaging betaamyloid plaques in human brain. Appl Radiat Isot 68: 22932297.
19. Wong DF, Rosenberg PB, Zhou Y, Kumar A, Raymont V, et al. (2010) In vivo
imaging of amyloid deposition in Alzheimer disease using the radioligand 18FAV-45 (Florbetapir F 18). J Nucl Med 51: 913920.
20. Hsiao IT, Huang CC, Hsieh CJ, Hsu WC, Wey SP, et al. (2012) Correlation of
early-phase 18F-florbetapir (AV-45/Amyvid) PET images to FDG images:
preliminary studies. Eur J Nucl Med Mol Imaging 39: 613620.
21. Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, et al.
(2002) Automated anatomical labeling of activations in SPM using a
macroscopic anatomical parcellation of the MNI MRI single-subject brain.
22. Huang KL, Lin KJ, Ho MY, Chang YJ, Chang CH, et al. (2012) Amyloid
deposition after cerebral hypoperfusion: evidenced on [18F]AV-45 positron
emission tomography. J Neurol Sci 319: 124129.
23. Hsieh CJ, Fan L, Lin KJ, Hsiao IT, Huang CC, et al. (2011) Precise F-18
florbetapir (AV-45) PET image spatial normalization using early-phase image
template. J Nucl Med 52: 549.
24. Ashburner J, Friston KJ (2005) Unified segmentation. NeuroImage 26: 839851.
25. Juh R, Kim J, Moon D, Choe B, Suh T (2004) Different metabolic patterns
analysis of Parkinsonism on the 18F-FDG PET. European Journal of Radiology
51: 223233.
26. Braak H, Braak E (1997) Frequency of Stages of Alzheimer-Related Lesions in
Different Age Categories. Neurobiol Aging 18: 351357.
27. Okello A, Koivunen J, Edison P, Archer HA, Turkheimer FE, et al. (2009)
Conversion of amyloid positive and negative MCI to AD over 3 years: an 11CPIB PET study. Neurology 73: 754760.
28. Small GW (2006) Diagnostic issues in dementia: neuroimaging as a surrogate
marker of disease. J Geriatr Psychiatry Neurol 19: 180185.
29. Herholz K, Ebmeier K (2011) Clinical amyloid imaging in Alzheimers disease.
The Lancet Neurology 10: 667670.
30. Hatashita S, Yamasaki H (2010) Clinically different stages of Alzheimers disease
associated by amyloid deposition with [11C]-PIB PET imaging. J Alzheimers
Dis 21: 9951003.
31. Barthel H, Senda M, Yamane T, Patt M, Sattler B, et al. (2010) Phase 1 trials on
florbetaben brain PET in elderly normal controls. J Nucl Med 51: 385.
32. Senda M, Yamane T, Sasaki M, Shimizu K, Barthel H, et al. (2010) Ethnic
comparability of safety and pharmacokinetics of florbetaben (BAY 94-9172) in
normal Caucasian and Asian subjects. J Nucl Med 51: 1754.
33. Fleisher AS, Chen K, Liu X, Roontiva A, Thiyyagura P, et al. (2011) Using
positron emission tomography and florbetapir F 18 to image cortical amyloid in
patients with mild cognitive impairment or dementia due to Alzheimer disease.
34. Jobst KA, Barnetson LPD, Shepstone BJ (1998) Accurate prediction of
histologically confirmed Alzheimers disease and the differential diagnosis of
dementia: the use of NINCDS-ADRDA and DSM-III-R criteria, SPECT, XRay CT, and Apo E4 in medial temporal lobe dementias. Int Psychogeriatr 10:
271302.

35. Ranginwala NA, Hynan LS, Weiner MF, White CLI (2008) Clinical criteria for
the diagnosis of Alzheimer disease: still good after all these years. Am J Geriatr
Psychiatry 16: 384-388.
36. LeVine III H, Walker LC (2010) Molecular polymorphism of Ab in Alzheimers
disease. Neurobiol Aging 31: 542548.
37. Lockhart A, Lamb JR, Osredkar T, Sue LI, Joyce JN, et al. (2007) PIB is a nonspecific imaging marker of amyloid-beta (Ab) peptide-related cerebral amyloidosis. Brain 130: 26072615.
38. Kemppainen NM, Aalto S, Wilson IA, Nagren K, Helin S, et al. (2007) PET
amyloid ligand [11C]PIB uptake is increased in mild cognitive impairment.
Neurology 68: 16031606.
39. Forsberg A, Engler H, Almkvist O, Blomquist G, Hagman G, et al. (2008) PET
40. Rowe CC, Ng S, Ackermann U, Gong SJ, Pike K, et al. (2007) Imaging betaamyloid burden in aging and dementia. Neurology 68: 17181725.
41. Aizenstein HJ, Nebes RD, Saxton JA, Price JC, Mathis CA, et al. (2008)
Frequent amyloid deposition without significant cognitive impairment among
the elderly. Arch Neurol 65: 15091517.
42. Rowe CC, Ellis KA, Rimajova M, Bourgeat P, Pike KE, et al. (2010) Amyloid
imaging results from the Australian Imaging, Biomarkers and Lifestyle (AIBL)
study of aging. Neurobiol Aging 31: 12751283.
43. Chetelat G, Villemagne VL, Bourgeat P, Pike KE, Jones G, et al. (2010)
Relationship between atrophy and beta-amyloid deposition in Alzheimer
disease. Ann Neurol 67: 317324.
44. Jack CR Jr, Lowe VJ, Weigand SD, Wiste HJ, Senjem ML, et al. (2009) Serial
PIB and MRI in normal, mild cognitive impairment and Alzheimers disease:
implications for sequence of pathological events in Alzheimers disease. Brain
132: 13551365.
45. Teipel SJ, Sabri O, Grothe M, Barthel H, Prvulovic D, et al. (2013) Perspectives
for Multimodal Neurochemical and Imaging Biomarkers in Alzheimers Disease.
J Alzheimers Dis 33: S329S347.
46. Roe CM, Mintun MA, DAngelo G, Xiong C, Grant EA, et al. (2008) Alzheimer
Disease and Cognitive Reserve: Variation of Education Effect With Carbon 11Labeled Pittsburgh Compound B Uptake. Arch Neurol 65: 14671471.
47. Barthel H, Luthardt J, Becker G, Patt M, Hammerstein E, et al. (2011)
Individualized quantification of brain b-amyloid burden: results of a proof of
mechanism phase 0 florbetaben PET trial in patients with Alzheimers disease
and healthy controls. Eur J Nucl Med Mol Imaging 38: 17021714.
48. Jack Jr CR, Knopman DS, Jagust WJ, Shaw LM, Aisen PS, et al. (2010)
cascade. The Lancet Neurology 9: 119128.
79
Multi-Method Analysis of MRI Images in Early

Diagnostics of Alzheimers Disease
Robin Wolz1., Valtteri Julkunen2., Juha Koikkalainen3, Eini Niskanen2,4, Dong Ping Zhang1, Daniel
Rueckert1, Hilkka Soininen2,5, Jyrki Lotjonen3*, the Alzheimers Disease Neuroimaging Initiative"
1 Biomedical Image Analysis Group, Department of Computing, Imperial College London, London, United Kingdom, 2 Department of Neurology, Kuopio University
Hospital, Kuopio, Finland, 3 Knowledge Intensive Services, VTT Technical Research Centre of Finland, Tampere, Finland, 4 Department of Applied Physics, University of
Eastern Finland, Kuopio, Finland, 5 Institute of Clinical Medicine, Neurology, University of Eastern Finland, Kuopio, Finland
Abstract
The role of structural brain magnetic resonance imaging (MRI) is becoming more and more emphasized in the early
diagnostics of Alzheimers disease (AD). This study aimed to assess the improvement in classification accuracy that can be
achieved by combining features from different structural MRI analysis techniques. Automatically estimated MR features
used are hippocampal volume, tensor-based morphometry, cortical thickness and a novel technique based on manifold
learning. Baseline MRIs acquired from all 834 subjects (231 healthy controls (HC), 238 stable mild cognitive impairment (SMCI), 167 MCI to AD progressors (P-MCI), 198 AD) from the Alzheimers Disease Neuroimaging Initiative (ADNI) database
were used for evaluation. We compared the classification accuracy achieved with linear discriminant analysis (LDA) and
support vector machines (SVM). The best results achieved with individual features are 90% sensitivity and 84% specificity
(HC/AD classification), 64%/66% (S-MCI/P-MCI) and 82%/76% (HC/P-MCI) with the LDA classifier. The combination of all
features improved these results to 93% sensitivity and 85% specificity (HC/AD), 67%/69% (S-MCI/P-MCI) and 86%/82% (HC/
P-MCI). Compared with previously published results in the ADNI database using individual MR-based features, the presented
results show that a comprehensive analysis of MRI images combining multiple features improves classification accuracy and
predictive power in detecting early AD. The most stable and reliable classification was achieved when combining all
available features.
Citation: Wolz R, Julkunen V, Koikkalainen J, Niskanen E, Zhang DP, et al. (2011) Multi-Method Analysis of MRI Images in Early Diagnostics of Alzheimers
Disease. PLoS ONE 6(10): e25446. doi:10.1371/journal.pone.0025446
Editor: Celia Oreja-Guevara, University Hospital La Paz, Spain
Received July 12, 2011; Accepted September 5, 2011; Published October 13, 2011
Copyright: 2011 Wolz et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
Funding: This project is partially funded under the 7th Framework Programme by the European Commission (http://cordis.europa.eu/ist/). Data collection and
sharing for this project was funded by the Alzheimers Disease Neuroimaging Initiative (ADNI; Principal Investigator: Michael Weiner; NIH grant U01 AG024904).
ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering (NIBIB), and through generous contributions
from the following: Pzer Inc., Wyeth Research, Bristol-Myers Squibb, Eli Lilly and Company, GlaxoSmithKline, Merck & Co. Inc., AstraZeneca AB, Novartis
Pharmaceuticals Corporation, Alzheimers Association, Eisai Global Clinical Development, Elan Corporation, plc, Forest Laboratories, and the Institute for the Study
of Aging, with participation from the U.S. Food and Drug Administration. Industry partnerships are coordinated through the Foundation for the National Institutes
of Health. The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimers Disease
Cooperative Study at the University of California San Diego. ADNI data are disseminated by the Laboratory of Neuro Imaging at the University of California Los
Angeles. Data used in the preparation of this article were obtained from the ADNI database (www.loni.ucla.edu/ADNI). As such, the investigators within the ADNI
contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. ADNI investigators
include (complete listing available at http://adni.loni.ucla.edu/wp-content/uploads/how_to_apply/ADNI_Authorship_List.pdf).
" For information on the Alzheimers Disease Neuroimaging Initiative please see the Acknowledgments section.
Introduction
The search for reliable biomarkers of AD-type pathology and

predictors of disease progression among MCI subjects is ongoing.
AD is characterized by neurofibrillary tangles and amyloid plaques
in the brain [5]. Degenerative changes in the human neurotransmitter system lead to atrophy in selected brain regions [6]. The
most promising candidate biomarkers are the ones derived from
structural and functional neuroimaging as well as those measured
in cerebrospinal fluid (CSF) and plasma [7]. Amyloid-based
measures like the CSF-peptide Ab42 and the uptake of the PiB
tracer on positron emission imaging (PET) show the earliest ADtype changes [7]. However, there is evidence that the number of
amyloid plaques reach their saturation levels already by the time
patients have clinically apparent symptoms of cognitive impairment [8,9], whereas atrophy, neuronal loss, synaptic loss, and the
number of tangles increase with severity of illness [10]. These

globally and one of the major healthcare issues of the future. It has
been estimated that during the next four decades the prevalence of
AD will quadruple from 27 to 106 million by which time 1 in 85
persons worldwide will be living with the disease [1]. Even a
modest delay of one year in disease onset and progression could
reduce the number of cases by 9 million [1]. Interventions are
postulated to be most effective when directed at patients at the
earliest stages of the disease, which underlines the importance of
early diagnosis of AD [2]. Mild cognitive impairment (MCI) is a
heterogeneous syndrome that increases the risk of developing AD
markedly [3]. However, not all MCI subjects convert to AD and
some may even return to normal cognition [4].
80

findings suggest that, although amyloid-based biomarkers may be

used as longitudinal markers of AD type pathology, they seem to
offer only limited insight into which MCI subjects will most likely
convert to AD in the near future. In a recently published dynamic
model of biomarker behavior in the AD spectrum, biomarkers
based on structural magnetic resonance imaging (MRI) have been
shown to be correlated with a progression from MCI to AD [11].
Such biomarkers could therefore improve the accuracy of early
AD diagnostics and reduce especially the amount of false positive
diagnoses. Besides providing chance for a more focused and earlier
intervention, structural MRI biomarkers of AD could also aid the
development of new disease-modifying drugs by acting as surrogate
markers of disease progression, reduce the number of subjects
needed to detect significant drug effect and provide quantitative
measures of treatment benefits [12].
It has been shown that the early diagnostics of AD can be
improved by using multiple different biomarkers simultaneously.
Usually these studies have combined MRI-based markers with
biomarkers based on positron emission tomography (PET) [13,14],
cerebrospinal fluid (CSF) [15,16] or both [1719]. Achieved
results vary from no additional benefit [15,17] to significant
improvement [13,14,16,20]. However, availability of all three
biomarkers (CSF, PET, MRI) is not very common in clinical
practice since obtaining all measures is laborious for the patient
and clinician, induces delays and increases the costs of the
diagnosis significantly. Furthermore, measurements obtained from
CSF and PET are considered invasive. Recent studies focusing on
only structural MRI have reached correct classification accuracys
(CCR) of 7694% in identifying healthy controls (HC) from
patients with AD and 6482% in predicting which MCI subjects
will convert to AD in the imminent future [2127]. The high
variation in these results can be attributed to differences in study
populations as well as evaluation designs. With the Alzheimers
Disease Neuroimaging Study (ADNI) [28], a large multi-center
study on MR imaging in AD has been established that is available
to the wider research community. Based on a large sub-group of
ADNI subjects, Cuingnet et al. [29] presented a comparison of ten
MRI-based feature extraction methods and their ability to
discriminate between clinically relevant subject groups. The ten
methods evaluated comprise five voxel-based methods, three
methods based on cortical thickness and two methods based on the
hippocampus. Best sensitivity/specificity values reported are 81%/
95% for AD vs HC, 70%/61% for S-MCI vs P-MCI and 73%/
85% for HC vs P-MCI.
In this paper we use the ADNI database to evaluate the ability
of the combination of different MR-based features to increase
classification accuracy. We evaluate the power of hippocampal
volume (HV), cortical thickness (CTH), tensor-based morphometry (TBM) and features extracted from a recently proposed
manifold-based learning (MBL) framework to discriminate healthy
controls from subjects with AD and to predict conversion from
MCI to AD. For evaluation we used all 834 ADNI baseline images
that were available from the ADNI webpage. Compared to
previous work this paper aims at establishing the improvement in
accuracy and stability that can be achieved by combining more
than one MR-based feature. To the best of our knowledge it is the
first comprehensive study that analyzes MRI-derived features for
the full ADNI dataset. For direct comparison with the work by
Cuingnet et al. [29] we also evaluated all results on the subset used
in their work.
To test the influence of the classification method used, we utilized
both support vector machines (SVMs) and a linear discriminant
analys (LDA) to evaluate classification accuracy (CCR), sensitivity
(SEN) and specificity (SPE) in each experiment.
Subjects
In the ADNI study, brain MR images were acquired at regular
intervals after an initial baseline scan from approximately 200
cognitively normal older subjects (HC), 400 subjects with mild
cognitive impairment (MCI), and 200 subjects with early AD.
Detailled inclusion/exclusion criteria used for the different subject
groups in ADNI are defined in [30]. The AD group has scores
between 2026 (inclusive) on the Mini-Mental State Examination
(MMSE) [31], and a Clinical Dementia Rating (CDR) [32] of 0.5
or 1.0. Furthermore, these subjects fulfil the NINCDS/ADRDA
criteria for probable AD [33]. MCI subjects included have MMSE
scores between 2430 (inclusive), a memory complaint, have
objective memory loss measured by education adjusted scores on
Wechsler Memory Scale Logical Memory II, a CDR of 0.5,
absence of significant levels of impairment in other cognitive
domains, essentially preserved activities of daily living, and an
absence of dementia [30]. Healthy subjects have MMSE scores
between 2430 (inclusive), a CDR of 0, are non-depressed, non
MCI, and nondemented. A more detailed description of the ADNI
study is given in Appendix S1.
All 834 ADNI subjects (231 HC, 238 S-MCI, 167 P-MCI, 198
AD) for which a 1.5T T1-weighted MRI scan at baseline was
available were included in this study. 167 subjects in the MCI
group converted to AD as of July 2011. We therefore
independently analysed progressive MCI (P-MCI) subjects and
subjects with a stable diagnosis of MCI (S-MCI).
Table 1 shows the demographics for the 834 study subjects.
Statistically significant differences in the demographics and clinical
variables between the study groups were assessed using Students
unpaired t-test. In this work, the difference was considered
statistically significant if pv0.05 if not stated otherwise. There
were more men than women in all other groups besides the AD
group. MMSE scores were significantly different in the pairwise
comparisons between all study groups. CDR scores of the HC and
AD groups are significantly different to the ones of the two MCI
groups. Healthy subjects had a significantly lower Geriatric
Depression Scale (GDS) compared to all other groups. Compared
to all other groups, AD subjects had significantly shorter education.
MRI Acquisition
Standard 1.5T screening/baseline T1-weighted images obtained using volumetric 3D MPRAGE protocol with resolutions
Table 1. Subjects.
Group
HC
S-MCI
P-MCI
AD
231
238
167
198
Men
52%
66%
62%
52%
Age
76.02 (5.0)
74.85 (7.8)
74.6 (7.0)
75.68 (7.7)
MMSE
29.1* (1.0)
27.3* (1.8)
26.6* (1.7)
23.3* (2.0)
CDR
0 (0)
0.49 (0.05)
0.50 (0)
0.75 (0.25)
GDS
0.83* (1.14)
1.60 (1.42)
1.53 (1.30)
1.67 (1.42)
Education
16.0 (2.8)
15.6 (3.1)
15.7 (2.9)
14.7* (3.1)
APOE4 status
(e3e4/e4e4)
23%/2%
31%/8%
50%/16%
42%/18%
Months to
conversion
18.2 (10.1)
*means statistically significant different from all other groups.

81
Instead of using just one template image, we used 30 randomly

selected images (10 controls, 10 MCIs, and 10 ADs) from the
ADNI database as template images. The template images were
used also in the classification analysis to maximize the number of
subjects. Each template image was registered to a study image, and
Jacobian maps were computed for each template image. To
combine the results of multiple templates, all template images were
registered to the mean anatomical template generated from the 30
images, and all the results were normalized to this reference space
[47]. The combination of the results was performed by averaging
the ROI-wise feature values of all the templates as described in
detail below.
Manifold-based learning. In this machine learning
approach, non-linear dimensionality reduction with Laplacian
eigenmaps [48] is used to learn features to discriminate between
different subject groups. Laplacian eigenmaps estimates the lowdimensional representation of a set of input images based on a
similarity graph that is defined with pairwise image similarities [48].
The hypothesis is that such a low-dimensional representation
captures the variability in the dataset in a more compact way than
pairwise image similarities directly. We estimate pairwise image
similarities from the intensity appearance in a region around
hippocampus and amygdala since both structures are known to be
affected by AD in an early stage. All images are aligned in a
template space using a coarse non-rigid registration (10 mm Bspline control-point spacing, [49]). Such a coarse non-rigid
alignment ensures that corresponding brain structures are aligned
but still allows to measure subject-specific differences. After
performing dimensionality reduction, the first 20 dimensions of
the resulting manifold are used as features to perform classification
with the different methods used. More details on the theory and
application of this manifold learning approach can be found in
[20,50]. Figure 1 exemplarily shows a 2D embedding of a set of
ADNI images acquired from healthy controls and subjects with AD.
It can be seen that even two embedding dimensions give a relatively
good separation between both groups. In our experiments we used a
higher dimensional space allowing better discrimination.
ranging from 0.9 mm|0.9 mm|1.20 mm to 1.3 mm|1.3 mm

|1.20 mm were included from the ADNI database. For detailed
information of the MRI protocols and preprocessing steps see [34].
Feature extraction
82
All fully automated feature extraction methods described below

were applied to images that were preprocessed by the ADNI
pipeline.
Hippocampal volume. Baseline hippocampal volumes
were measured using an approach based on fast and robust
multi-atlas segmentation [35,36]. In this approach, multi-atlas
label propagation is applied in combination with atlas selection to
obtain the hippocampus segmentation. A set of hippocampus
atlases is selected from a pool of atlas images according to image
similarity with the query image. After registering all atlases to the
query image, a spatial prior is generated from the multiple label
maps. This spatial prior is then used to obtain a final segmentation
based on an expectation maximization (EM) segmentation
algorithm [37].
Cortical thickness. CTH is measured in the baseline
T1-weighted structural MR images by using an automated
computational surface-based method developed at the McConnell
Brain Imaging Centre, Montreal Neurological Institute, McGill
University, Montreal, Canada (http://www2.bic.mni.mcgill.ca/)
[38]. Individual MRI volumes were registered to standard space
using the ICBM152 template [39]. Intensity non-uniformities were
corrected [40] before the final brain mask was calculated [41].
Tissues were segmented into white matter (WM), grey matter (GM)
and cerebrospinal fluid (CSF) using the INSECT-algorithm [42]
and the magnitude of PVE was estimated by using the trimmed
minimum covariance determinant (TMCD) method [43]. The
brains were divided automatically into two separate hemispheres
and the inner and outer surfaces of the cortex were extracted
according to intersections between WM and GM (white matter
surface, WMS) as well as GM and CSF (grey matter surface, GMS)
using the Constrained Laplacian-Based Automated Segmentation
with Proximities (CLASP) algorithm [44]. The inner surface was
first formed by deforming an ellipsoid polygon mesh to the shape of
the WMS. GMS was obtained by further expanding the inner
surface. Each polygon mesh surface consisted of 81,920 polygons
and 40,962 nodes per hemisphere. The thickness of the cortex was
defined at each linked node as the distance between the two
concentrically linked polygon meshes on the WMS and the GMS.
This t-link metric has been proven to be the simplest yet most
precise way to determine cortical thickness [38]. Although MR
images were transformed to standard space to allow for group
analysis, thickness calculations were performed in each subjects
native space. Finally, cortical thickness maps were smoothed with a
20 mm FWHM diffusion smoothing kernel to improve the signalto-noise ratio and statistical power [45]. The described toolbox did
not achieve satisfactory results on some study subjects because of i)
failure in tissue segmentation and brain masking (48 subjects) and ii)
failure in partial volume effect estimation (59 subjects). As a result
the pipeline crashed and CTH measures were not obtained for 76
subjects (24 control, 35 MCI, 17 AD). Also the cortical model of 31
subjects (10 control, 13 MCI, 8 AD) was completely deformed and
thus unusable. For these 107 subjects the CTH features were
considered as missing values. CTH features used in the classification
experiments are introduced below.
Tensor-based morphometry. The TBM analysis was
performed using a multi-template approach [46,47]. In TBM, a
template image is non-rigidly registered to a study image, and,
typically, the determinant of the Jacobian matrix (the Jacobian) of
the deformation is used to measure the voxel-level morphometry.
ROI-wise features for CTH and TBM

Both CTH and TBM analyses produce local (point-wise)
information, either on cortical thickness or the volume. Thus,
the number of original features is enormous, and to make the
classification more efficient and robust, the number of features has
to be reduced. We evaluated both features in a statistical region of
interest (ROI) defined as detailed in Appendix S2. Figures 2 and 3
show t-values for statistically significant differences between study
groups for TBM and CTH respectively. A detailed description of
the definition of these statistical ROIs is given in Appendix S2.
Study design
Table 2 presents an overview on the features calculated for all
834 available ADNI baseline images. All feature values were
corrected for age and gender using a linear regression model
where control subjects were used as the training set, i.e., the
normal, not disease-related, age and gender related differences in
the classification features were removed. Feature selection was
then carried out on the corrected feature sets using stepwise
regression [51].
We used two subsets to perform classification:
I.
II.
All 834 available baseline images described in the subjects

section
509 baseline images used by Cuingnet et al. [29] and detailed
in their publication.

Figure 1. 2D manifold embedding of a set of images acquired from healthy controls (red) and subjects with AD (blue).
Consequently, in overall, the classification evaluation was

performed using 300 (3|100) repetitions, and the results
presented in this paper are the average values of all these
classifications.
Dataset II. Statistical ROIs for CTH and TBM feature
extraction were calculated from the 325 baseline images that are
not part of dataset II. In order to allow direct comparison of
classification accuracy with the work by Cuingnet et al. [29],
separate training and testing sets for the different comparisons
were defined using the exact sub-groups reported in their
manuscript. Around 50% of all subjects are used to train the
different types of classifiers and the reported results are based on
classifying the remaining subjects.
The following sections describe the definition of the statistical

ROIs and evaluation strategy used for the two datasets
respectively.
Dataset I. In order to perform the study using crossvalidation in the full dataset, it was divided into three equally
sized parts. One part was used to perform the statistical tests for
the CTH and TBM features, and the remaining two parts were
used to evaluate the classification accuracy. This was repeated
three times so that each part was once used to perform the
statistical tests. Afterwards, the results of the three repetitions were
averaged. The classification accuracy was evaluated using leave-Nout cross validation on those subjects not included in the statistical
tests. Five percent of the evaluation subjects were regarded as the
test set, and the remaining 95% of the subjects were used to train a
classifier which was then applied to the test set. This was repeated
table-1-caption100 times, each time selecting randomly the test set
subjects. Finally, the results of the 100 repetitions were averaged.
Classification methods
We used two different widely used methods to perform
classification based on individual features and their combination:
Figure 2. Results for voxelwise t-tests for statistically significant group differences with features extracted from TBM.
83
Figure 3. Results for t-tests for statistically significant group differences based on cortical thickness measurements.
Linear discriminant analysis (LDA). Linear discriminant

analysis (LDA) is a widely used technique to find a linear
combination of features to best separate several classes [52]. In this
work we used LDA as implemented in the classify function in
Matlab with a multivariate normal density model with uninformative priors (p = 0.5).
Support vector machines (SVM). Support vector machines
use training data to find a separating hyperplane in the ndimensional training space that best separates two subject groups
[53]. Test subjects are then classified according to their position
relative to the defined hyperplane in the n-dimensional feature
space. We used the libSVM library to perform the analysis. The
radial basis function kernel was selected based on the guidelines
provided by the libSVM library (Software available 2.3.2011 at
http://www.csie.ntu.edu.tw/cjlin/libsvm).
significant improvements achieved when combining all features are

marked with { (pv0.0001). To test for significance, unpaired t-tests
were carried out between distribution estimates for the corresponding classification rates based on the multiple runs. All estimated
distributions passed a normality test using a Kolmogorov-Smirnov
test at a~0:05.
For direct comparison with work presented by Cuingnet et al.
[29], we performed classification based on the training- and testing
sets defined in their manuscript as described above. S-MCI and PMCI groups are defined in the same way as in the original
publication. Sensitivity and specificity values for the classification
in all three clinical pairings are reported in Table 6. Following the
clear advantage for LDA in the performance on the full dataset,
we only report results with this classifier for dataset II.
Discussion
Results
In this study we assessed the automatic diagnostic capabilities of

4 structural MRI features (MBL, HC, CTH, TBM) separately and
combined in 834 baseline images acquired in the ADNI study.
When applied separately, TBM provided the overall best results,
closely followed by MBL. Combining all features improved
the results in all study experiments. Our results show how a
combination of different MRI-based features can improve results
based on only one measurement, resulting in a more powerful and
stable classifier. The most significant improvement of the combination
We used both classification methods to measure classification

accuracy based on individual features as well as the combination of
all features. The results for the comparisons HC vs AD, HC vs PMCI and S-MCI vs P-MCI in the full ADNI database are presented
in Tables 3, 4 and 5 respectively. Presented are classification
accuracy (CCR), sensitivity (SEN) and specificity (SPE). Furthermore, the 95% confidence interval for the classification accuracy is
estimated based on the multiple classification runs. Statistically
Table 2. Features used in the study.
Method
No of features
Description
Hippocampal volume (HV)
total volume of left and right hippopcampus

average cortical thickness within a ROI defined based on group-level statistical analysis
Cortical thickness
9 (HC vs AD)
(CTH)
7 (HC vs P-MCI)
Tensor-based morphometry (TBM)
84
average Jacobian of atrophic voxels within a ROI, weighted based on voxel-wise p-values
Manifold-based learning (MBL)
20
coordinates of a subject in a low-dimensional manifold space learned from pairwise image

similarities
8 (S-MCI vs P-MCI)
84

Table 3. Classification results for HC vs AD.
Feature
LDA
Table 5. Classification results for S-MCI vs P-MCI.
SVM
Feature
CCR [95% CI]
SEN
SPE
CCR [95% CI]
SEN
SPE
MBL
85{ [64 100]
87
83
85 [64 100]
87
83
HV
81{ [57 100]
81
79
81{ [57 100]
84
77
CTH
81{ [64 100]
89
71
82{ [57 100]
90
TBM
87{ [71 100]
90
84
87 [71 100]
89
All
89 [71 100]
93
85
86 [71 100]
94
SEN
SPE
CCR [95% CI]
MBL
65{ [36 86]
64
66
65{ [43 86]
77
48
HV
65{ [36 86]
63
67
62 [36 86]
83
33
73
CTH
56{ [29 86]
63
45
59 [36 79]
96
03
84
TBM
64{ [36 86]
65
62
64{ [36 86]
77
44
78
All
68 [43 93]
67
69
60 [36 86]
92
14
SEN
SPE

pv0.0001.
over the best individual feature can be observed for HC vs P-MCI

with 5% units followed by 3 and 2% units for S-MCI vs P-MCI and
HC vs AD, respectively. These improvements lead to 20, 12 and 9
subjects more being correctly classified respectively when using the
combined feature set as compared to the best single feature for every
comparison. Comparing two classification approaches based on LDA
and SVMs resulted in a clear advantage of the former.
Several studies reported classification results using single MRI
methods for the HC/AD classification (Table 7). Liu et al. [24]
reported SEN/SPE of 92/90 in the classification of HC/AD
subjects using regional cortical volumes in the AddNeuroMed
dataset. McEvoy et al. [26] report a CCR of 89 on images from
the ADNI database using features from cortical thickness and
structural volumes. Vemuri et al. [54] present a SEN/SPE of 86/
86 on 380 subjects using the STAND score. In our study the
results obtained with single methods are lower (7190) but almost
identical when the methods were combined. It should be noted,
however, that Liu and colleagues did not use cross-validation or
separate training/testing sets when producing the results which
could lead to overestimation of the results in a dataset outside the
study cohort. Gerardin et al. [23] acquired a high SEN/SPE of
96/92 by using hippocampal shape analysis, but the number of
subjects (25 HC, 23 AD) was quite low in order to produce results
with good generalizability. Westman et al. [55] reported a CCR of
82 for HC vs AD classification and 73 for HC vs P-MCI
classification by using various regional brain volumes. Our results
are substantially more accurate, the group sizes are larger and
clinical follow-up time is one year longer. Chupin et al. [21]
reported SEN/SPE of 75/77 (hippocampal volume) and Querbes
et al. [27] a CCR of 85 (cortical thickness), both lower than the
results acquired with the combination of features or TBM features

independently in our study.
Varying results concerning AD prediction (S-MCI/P-MCI
classification using baseline measurements) have been published
(Table 7): Querbes et al. [27] reported a CCR of 73, Liu et al. [25]
a SEN/SPE of 76/68, Chupin et al. [21] reported a SEN/SPE of
60/65 and Davatzikos et al. [15] SEN/SPE of 95/38. Our results
with separate and combined baseline features lie in the range of
these results (SEN/SPE 63/67, 64/66 and 67/69 when using HV,
MBL and the combined features, respectively).
There can be several explanations for the variation in the
reported results. A majority of the studies in this field have used
different statistical methods and MRI feature extraction strategies
on different datasets, which makes a comparison of the results
complicated. Also the variation in the size of the study samples and
the use (or ignoring) of cross-validation or separate training/testing
sets are important factors, which both have crucial impact on the
reliability and generalizability of the results. In Lotjonen et al.
[36], we demonstrated that choosing from a population of 350
cases several times 2/3 for the training set and 1/3 for the test set
and using hippocampus volume as a classification feature can lead
to any classification accuracy between 53% and 77%. This
observation is also confirmed by the high confidence intervals for
the classification accuracies reported in Tables 3, 4 and 5. This
shows that a fair comparison of methods based on the classification
accuracy is difficult if not exactly the same data and classification
approaches are used. Furthermore, since the ADNI study is still
ongoing, several subjects labeled as S-MCI will progress in the
future to the P-MCI group.
A recent study with a subset of ADNI subjects assessed the
classification performance of several structural MRI methods in
experiments comparable to our investigation [29]. Reported
Table 4. Classification results for HC vs P-MCI.
LDA
SVM
CCR [95% CI]

pv0.0001. CCR = Correct classification rate, SEN = Sensitivity, SPE = Specificity.
Feature
LDA
Table 6. Classification results based on a subset of ADNI that

was previously used for classification by Cuingnet et al. [29].
SVM
CCR [95% CI]
SEN
SPE
CCR [95% CI]
SEN
SPE
MBL
78{ [54 100]
81
75
77{ [54 92]
84
69
HV
76{ [54 92]
77
76
78 [54 92]
83
71
CTH
77{ [54 100]
85
65
77 [54 100]
89
62
TBM
79{ [62 100]
82
76
80{ [62 100]
85
74
All
84 [62 100]
86
82
82 [62 100]
93
67
Feature
SEN

pv0.0001.
HC vs AD
SPE
HC vs P-MCI
S-MCI vs P-MCI
SEN
SEN
SPE
SPE
MBL
90
74
84
92
55
76
HV
80
69
75
76
63
70
CTH
85
75
86
59
72
35
TBM
93
76
90
84
63
59
All
94
76
94
89
69
54
85
Table 7. Classification results of healthy control (HC), mild cognitive impairment (MCI) and Alzheimers disease subjects reported
in the recent literature.
Study
Features
HC vs AD
HC vs P-MCI
S-MCI vs P-MCI
CCR
SEN
SPE
CCR
SEN
SPE
CCR
SEN
SPE
Liu et al. [24]
333
Cortical volumes
91
92
90
Gerardin et al. [23]*
70
Hippocampus shape
94
96
92
Chupin et al. [21]*
605
Hippocampus volume
76
75
77
64
60
65
Querbes et al. [27]*
382
Cortical thickness
85
73
75
68
Liu et al. [25]
312
Amygdala/caudate volumes
69
76
68
Davatzikos et al. [15]*
356
SPARE-AD index
56
95
38
Cuingnet et al. [29]*
509
Various
81
95
73
85
62
69
Hinrichs et al. [14]*
159
MRI & PET
81
60
92
14
Westman et al. [55]
351
Various volumes
82
73
McEvoy et al. [26]*
398
Cortical thickness/various volumes
89
83
93
Vemuri et al. [54]
380
STAND score
86
86
N = Number of study subjects,

* = ADNI dataset.
86
Considering solely the classification accuracies of the present

study and those reported in literature, it seems questionable if the
collection of several biomarkers is worth the effort and resource. A
combination of different features extracted from a single MRI
seems to provide results that are comparable or better than those
obtained with other or multiple biomarkers. In a clinical point of
view, this is interesting since it means that a single MRI scan
provides not only aid to differential diagnostics of cognitive
impairment, but also reliably describes a persons phase in the HC/
AD continuum. MRI is also widely available, non-invasive and
often useful in the differential diagnostics of memory problems
thus making it a compelling option as the first biomarker that
would be obtained from a patient with mild memory problems.
However, a comprehensive differential diagnostics between AD
and non-AD cognitive impairments will still require assessment of
various different biomarkers. Also, it should be noted that the
computational techniques used in this paper are not widely
available in the clinical environment and thus limit their usage in
the clinical work at present.
Strengths of the presented study are i) the use of multiple
features extracted from one imaging modality, ii) large groups, iii)
rigorous validation process of the results using cross-validation,
and iv) results comparable or better than the ones published so far.
Our study has also some limitations that should be mentioned.
The results are obtained from a single (although collected from
multiple sites) cohort and should be also validated in other cohorts.
A longer clinical follow-up time would be needed to see if the
classification results of S-MCI/P-MCI experiment changed when
more of the MCI subjects converted to AD. Furthermore, the
ADNI study does not provide postmortem pathological confirmation of the clinical status. With this limitation, individual subjects
might be wrongly categorized. Although a rigorous validation
process was used, optimally we need to establish standardized cutoffs that would be well generalizable to other cohorts outside
ADNI. That is, however, beyond the possibilities of this study and
will require vast standardization and validation procedures. Also,
the CTH pipeline had problems especially with severely atrophied
brains or MRI scans with poor image quality. A more robust
pipeline would be desirable in order to guarantee a more reliable
feature extraction.
SEN/SPE lie in the ranges 59/8181/95 (HC vs AD) and 70/73

73/85 (HC vs P-MCI). While most methods tested did not exceed
the accuracy of a random classifier for the discrimination between
S-MCI and P-MCI, the best results reported for this task were a
SEN/SPE of 62/69 when using hippocampal volume. To allow a
direct comparison of the results reported by Cuingnet et al. [29],
we evaluated our features on the exact same training- and testing
sets used in their paper. This direct comparison shows that our
results compare favourably to other, established methods in
neuroimaging. For HC vs AD classification, individual features in
our study give more sensitive but less specific results than most
methods in the previous publication. Combining all features gives
an overall better classification accuracy than the majority of
previously tested methods. Our results on the combined feature set
furthermore outperform the majority of methods tested by
Cuingnet et al. [29] when predicting MCI conversion as well as
all methods for the classification between HC and P-MCI. A
significant difference in classification accuracy can be observed
between the full ADNI dataset and this smaller subset used for
comparison with previous work. Reasons may include a strict
separation into trainin- and testing sets which may result in less
generalisability as well as the shorter follow-up period that was
considered to define progression to AD.
Some studies have also combined different biomarkers (CSF,
MRI, PET) with the idea of measuring different aspects of AD
pathology and thus improve the classification accuracy. Hinrichs
et al. [14] improved their HC/AD classification CCR by a few %
units to 81 by combining MRI and PET. Eckerstrom et al. [16]
studied the separation of a unified HC/S-MCI group from P-MCI
group with CSF proteins and manual hippocampal volumes. They
found CSF to be superior to MRI (SEN/SPE of 95/79 vs 86/66)
while the combination performed best (SEN/SPE 90/91).
However, it should be noted that the study sample in that
particular study was small (a total of 68 subjects) and neither crossvalidation or separate training/testing sets were used in order to
ensure good generalizability of the results. In Kohannim et al.
[17], the improvement from using multiple biomarkers was not
significant and Davatzikos et al. [15] reported marginal improvements which, however, may be related to the fact that results with
only one biomarker were not very good to begin with.

ADNI contributed to the design and implementation of ADNI and/or

provided data but did not participate in analysis or writing of this report.
ADNI investigators include (complete listing available at http://adni.loni.
ucla.edu/wp-content/uploads/how_to_apply/ADNI_Authorship_List.
pdf).
Appendix S1
The Alzheimers Disease Neuroimaging Initiative.
(DOCX)
Appendix S2 ROI-wise features for CTH and TBM.
(DOCX)
Acknowledgments
Conceived and designed the experiments: RW VJ JK DR HS JL.

Performed the experiments: RW VJ JK EN DPZ. Analyzed the data: RW
VJ JK. Contributed reagents/materials/analysis tools: RW JK DR JL.
Wrote the paper: RW VJ.
Data used in the preparation of this article were obtained from the ADNI
database (www.loni.ucla.edu/ADNI). As such, the investigators within the
References
24. Liu Y, Paajanen T, Zhang Y, Westman E, Wahlund LO, et al. (2009)
in differentiating AD, MCI and control groupsThe AddNeuroMed study.
Neurobiology of Aging In Press, Corrected Proof: -.
25. Liu Y, Paajanen T, Zhang Y, Westman E, Wahlund LO, et al. (2010) Analysis of
regional MRI volumes and thicknesses as predictors of conversion from mild
13751385.
26. McEvoy L, Fennema-Notestine C, JC R, Hagler D, Holland D, et al. (2009)
Alzheimer disease: quantitative structural neuroimaging for detection and
prediction of clinical and structural changes in mild cognitive impairment.
Radiology 251: 1950205.
27. Querbes O, Aubry F, Pariente J, Lotterie JA, Dmonet JF, et al. (2009) Early
diagnosis of alzheimers disease using cortical thickness: impact of cognitive
reserve. Brain 132: 20362047.
28. Mueller SG, Weiner MW, Thal LJ, Petersen RC, Jack C, et al. (2005) The
Alzheimers Disease Neuroimaging Initiative. Neuroimaging Clinics of North
America 15: 869877.
29. Cuingnet R, Gerardin E, Tessieras J, Auzias G, Lehricy S, et al. (2010)
Automatic classification of patients with Alzheimers disease from structural
MRI: A comparison of ten methods using the ADNI database. NeuroImage In
Press, Corrected Proof: -.
30. Petersen RC, Aisen PS, Beckett LA, Donohue MC, Gamst AC, et al. (2010)
Alzheimers Disease Neuroimaging Initiative (ADNI): clinical characterization.
31. Folstein MF, Folstein SE, McHugh PR (1975) Mini-mental state: A practical
method for grading the cognitive state of patients for the clinician. Journal of
Psychiatric Research 12(3): 189198.
32. Morris J (1993) The Clinical Dementia Rating (CDR): current version and
The Alzheimers disease neuroimaging initiative (ADNI): MRI methods. Journal
of Magnetic Resonance Imaging 27: 685691.
35. Lotjonen JM, Wolz R, Koikkalainen JR, Thurfjell L, Waldemar G, et al. (2010)
Fast and robust multi-atlas segmentation of brain magnetic resonance images.
36. Lotjonen J, Wolz R, Koikkalainen J, Julkunen V, Thurfjell L, et al. (2011) Fast
and robust extraction of hippocampus from mr images for diagnostics of
alzheimers disease. NeuroImage 56: 185196.
37. Van Leemput K, Maes F, Vandermeulen D, Suetens P (1999) Automated
model-based tissue classification of MR images of the brain. IEEE Transactions
on Medical Imaging 18: 897908.
38. Lerch J, Evans A (2005) Cortical thickness analysis examined through power
analysis and a population simulation. NeuroImage 24: 163173.
39. Mazziotta J, Toga A, Evans A, Fox P, Lancaster J, et al. (2001) A probabilistic
atlas and reference system for the human brain: International Consortium for
Brain Mapping (ICBM). Philos Trans R Soc Lond B Biol Sci 356: 12931322.
automatic correction of intensity nonuniformity in MRI data. IEEE Transactions on Medical Imaging 17: 8797.
143155.
42. Zijdenbos AP, Forghani R, Evans AC (1998) Automatic quantification of MS
lesions in 3D MRI brain data sets: Validation of INSECT. In: III WMW,
Colchester ACF, Delp SL, eds. MICCAI Springer, volume 1496 of Lecture
Notes in Computer Science. pp 439448.
43. Tohka J, Zijdenbos A, Evans A (2004) Fast and robust parameter estimation for
statistical partial volume models in brain mri. NeuroImage 23: 8497.
44. Kim JS, Singh V, Lee JK, Lerch J, Ad-Dabbagh Y, et al. (2005) Automated 3-d
extraction and evaluation of the inner and outer cortical surfaces using a
laplacian map and partial volume effect classification. NeuroImage 27: 210221.
45. Chung M, Taylor J (2004) Diffusion smoothing on brain surface via finite
element method. In: ISBI IEEE. pp 432435.
1. Brookmeyer R, Johnson E, Ziegler-Graham K, Arrighi HM (2007) Forecasting

the global burden of Alzheimers disease. Alzheimers and Dementia 3: 186191.
2. Cummings J, Doody R, Clark C (2007) Disease-modifying therapies for
alzheimer disease: challenges to early intervention. Neurology 69: 16221634.
3. Petersen R (2001) Practice parameter: early detection of dementia: mild
cognitive impairment (an evidence-based review). Neurology 56: 1133.
4. Gauthier S, Reisberg B, Zaudig M, Petersen R, Ritchie K, et al. (2006) Mild
cognitive impairment. Lancet 367: 12621270.
5. Braak H, Braak E (1991) Neuropathological stageing of Alzheimer-related
changes. Acta Neuropathologica 82: 239259.
6. Wenk G (2003) Neuropathologic changes in alzheimers disease. Journal of
Clinical Psychiatry 64 Suppl 9.
7. Hampel H, Brger K, Teipel SJ, Bokde AL, Zetterberg H, et al. (2008) Core
candidate neurochemical and imaging biomarkers of alzheimers disease.
Alzheimers and Dementia 4: 3848.
8. Hyman BT, Marzloff K, Arriagada PV (1993) The lack of accumulation of senile
plaques or amyloid burden in Alzheimers disease suggests a dynamic balance
between amyloid deposition and resolution. J Neuropathol Exp Neurol 52:
594600.
9. Gmez-Isla T, Hollister R, West H, Mui S, Growdon J, et al. (1997) Neuronal loss
correlates with but exceeds neurofibrillary tangles in alzheimers disease. Ann
Neurol 41: 1724.
10. Ingelsson M, Fukumoto H, Newell KL, Growdon JH, Hedley-Whyte ET, et al.
(2004) Early abeta accumulation and progressive synaptic loss, gliosis, and tangle
formation in ad brain. Neurology 62: 92531.
11. Jack C, Jr., Knopman D, Jagust W, Shaw L, Aisen P, et al. (2010) Hypothetical
model of dynamic biomarkers of the alzheimers pathological cascade. Lancet
12. Hampel H, Frank R, Broich K, Teipel S, Katz J, abd Hardy RG, et al. (2010)
Biomarkers for alzheimers disease: academic, industry and regulatory
perspectives. NatRevDrug Discov 9: 560574.
biomarkers of prodromal alzheimers disease: A high-dimensional pattern
classification study. NeuroImage 41: 277285.
14. Hinrichs C, Singh V, Xu G, Johnson S (2009) Mkl for robust multi-modality ad
classification. In: MICCAI (II) Springer, volume 5762 of Lecture Notes in
Computer Science. pp 786794.
classification. Neurobiology of Aging In Press, Corrected Proof: -.
16. Eckerstrom C, Andreasson U, Olsson E, Rolstad S, Blennow K, et al. (2010)
Combination of hippocampal volume and cerebrospinal uid biomarkers
improves predictive value in mild cognitive impairment. Dement Geriatr Cogn
Disord 29: 294300.
17. Kohannim O, Hua X, Hibar DP, Lee S, Chou YY, et al. (2010) Boosting power
for clinical trials using classifiers based on multiple biomarkers. Neurobiology of
Aging 31: 14291442.
18. Landau S, Harvey D, Madison C, Reiman E, Foster N, et al. (2010) Comparing
predictors of conversion and decline in mild cognitive impairment. Neurology
75: 2308.
19. Walhovd KB, Fjell AM, Brewer J, McEvoy LK, Fennema-Notestine C, et al.
(2010) Combining MR Imaging, Positron-Emission Tomography, and CSF
Biomarkers in the Diagnosis and Prognosis of Alzheimer Disease. American
Journal of Neuroradiology 31: 347354+.
20. Wolz R, Aljabar P, Hajnal JV, Lotjonen J, Rueckert D (2011) Manifold learning
combining imaging with non-imaging information. In: IEEE International
Symposium on Biomedical Imaging. pp 960963.
21. Chupin M, Hammers A, Liu R, Colliot O, Burdett J, et al. (2009) Automatic
segmentation of the hippocampus and the amygdala driven by hybrid
constraints: Method and validation. NeuroImage 46: 749761.
22. Devanand DP, Pradhaban G, Liu X, Khandji A, De Santi S, et al. (2007)
Hippocampal and entorhinal atrophy in mild cognitive impairment - Prediction
of Alzheimer disease. Neurology 68: 828836+.
Alzheimers disease and mild cognitive impairment from normal aging.
87
46. Brun CC, Lepor N, Pennec X, Lee AD, Barysheva M, et al. (2009) Mapping the
regional inuence of genetics on brain structure variability a tensor-based
morphometry study. NeuroImage 48: 3749.
47. Koikkalainen J, Lotjonen J, Thurfjell L, Rueckert D, Waldemar G, et al. (2011)
Multi-template tensor-based morphometry: Application to analysis of alzheimers disease. NeuroImage 56: 11341144.
48. Belkin M, Niyogi P (2003) Laplacian eigenmaps for dimensionality reduction
and data representation. Neural Computation 15: 13731396.
49. Rueckert D, Sonoda LI, Hayes C, Hill DLG, Leach MO, et al. (1999) Nonrigid
registration using free-form deformations: Application to breast MR images.
IEEE Transactions on Medical Imaging 18: 712721.
Measurement of hippocampal atrophy using 4D graph-cut segmentation:
Application to ADNI. NeuroImage 52: 10091018.
51. Draper NR, Smith H (1998) Applied Regression Analysis (Wiley Series in
Probability and Statistics). Wiley.
52. Krzanowski WJ (1988) Principles of Multivariate Analysis: A Users Perspective
Oxford University Press.
53. Cortes C, Vapnik V (1995) Support-vector networks. Machine Learning 20:
273297.
54. Vemuri P, Gunter JL, Senjem ML, Whitwell JL, Kantarci K, et al. (2008)
Alzheimers disease diagnosis in individual subjects using structural mr images:
Validation studies. NeuroImage 39: 11861197.
Multivariate analysis of mri data for alzheimers disease, mild cognitive
impairment and healthy controls. NeuroImage 54: 11781187.
88
Combined Evaluation of FDG-PET and MRI Improves

Detection and Differentiation of Dementia
Juergen Dukart1*, Karsten Mueller1, Annette Horstmann1, Henryk Barthel3, Harald E. Moller1, Arno
Villringer1,2, Osama Sabri3, Matthias L. Schroeter1,2
1 Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, 2 Day Clinic of Cognitive Neurology, University of Leipzig, Leipzig, Germany,
3 Department of Nuclear Medicine, University of Leipzig, Leipzig, Germany
Abstract
Introduction: Various biomarkers have been reported in recent literature regarding imaging abnormalities in different types
of dementia. These biomarkers have helped to significantly improve early detection and also differentiation of various
dementia syndromes. In this study, we systematically applied whole-brain and region-of-interest (ROI) based support vector
machine classification separately and on combined information from different imaging modalities to improve the detection
and differentiation of different types of dementia.
Methods: Patients with clinically diagnosed Alzheimers disease (AD: n = 21), with frontotemporal lobar degeneration (FTLD:
n = 14) and control subjects (n = 13) underwent both [F18]fluorodeoxyglucose positron emission tomography (FDG-PET)
scanning and magnetic resonance imaging (MRI), together with clinical and behavioral assessment. FDG-PET and MRI data
were commonly processed to get a precise overlap of all regions in both modalities. Support vector machine classification
was applied with varying parameters separately for both modalities and to combined information obtained from MR and
FDG-PET images. ROIs were extracted from comprehensive systematic and quantitative meta-analyses investigating both
disorders.
Results: Using single-modality whole-brain and ROI information FDG-PET provided highest accuracy rates for both,
detection and differentiation of AD and FTLD compared to structural information from MRI. The ROI-based multimodal
classification, combining FDG-PET and MRI information, was highly superior to the unimodal approach and to the wholebrain pattern classification. With this method, accuracy rate of up to 92% for the differentiation of the three groups and an
accuracy of 94% for the differentiation of AD and FTLD patients was obtained.
Conclusion: Accuracy rate obtained using combined information from both imaging modalities is the highest reported up
to now for differentiation of both types of dementia. Our results indicate a substantial gain in accuracy using combined
FDG-PET and MRI information and suggest the incorporation of such approaches to clinical diagnosis and to differential
diagnostic procedures of neurodegenerative disorders.
Citation: Dukart J, Mueller K, Horstmann A, Barthel H, Moller HE, et al. (2011) Combined Evaluation of FDG-PET and MRI Improves Detection and Differentiation of
Dementia. PLoS ONE 6(3): e18111. doi:10.1371/journal.pone.0018111
Editor: Wang Zhan, University of California, San Francisco, United States of America
Received August 30, 2010; Accepted February 25, 2011; Published March 23, 2011
Copyright: 2011 Dukart et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
Funding: The work was funded by the Max Planck Society. The funders had no role in study design, data collection and analysis, decision to publish, or
preparation of the manuscript.
problematic. For this differentiation, accuracy rates ranging

between 84 and 89% are still in need of improvement, especially
due to a substantially lower sensitivity compared with specificity of
actual methods [10],[12],[15]. Nevertheless, the use of biomarkers
has significantly helped to improve diagnostic accuracy compared
with diagnoses based solely on clinical and neuropsychological
evaluation [16],[17]. For these reasons, recent studies have
suggested to incorporate imaging findings into criteria for
diagnosis of dementia [17],[18].
For AD patients imaging studies have shown reduced glucose
consumption mainly in parietotemporal and posterior cingulate
cortices [9],[20],[21] and structural changes in the hippocampus
and entorhinal area relative to healthy controls [9],[21],[22]. In
FTLD patients, atrophy and reduced metabolic rate for glucose
Introduction
In recent research, various biomarkers have been reported to
differentiate between early stages of dementia and healthy control
subjects or between different types of neurodegenerative disorders,
suggesting an integration of these would improve diagnostic
accuracy of dementia [1][9].
For the detection of dementia, accuracy rates significantly above
90% have recently been reported using univariate and multivariate statistical approaches in magnetic resonance imaging (MRI)
and [F18]fluorodeoxyglucose positron emission tomography
(FDG-PET) [1],[10][14]. However, the differentiation of the
two most common types of dementia, namely Alzheimers disease
(AD) and frontotemporal lobar degeneration (FTLD), is still
89
Multimodal Imaging in Dementia
have been reported to be predominately located in the medial

thalamus, amygdala and in frontotemporal and anterior cingulate
cortices [4],[8],[20],[21],[23].
For multivariate differentiation of different types of dementia
support vector machine classification (SVM) is used based on
whole-brain voxel information [10] or most frequently on ROI
values [6],[12][13],[24][26]. A major problem of the ROIbased approach is the limited generalizability of the trained
classifier, because the ROIs are selected based on features showing
a between-group differentiation in the same groups in a univariate
analysis. Although ROIs selected with this method provide a good
discrimination between groups used in these specific studies, they
might show significantly reduced discrimination power when
applied to new data sets. This could be the case if the selected
regions just detect differences between groups, which are not
necessarily attributed to the specific neurodegenerative disorder.
Furthermore, AD and FTLD patients have been shown to develop
a differential regional pattern of glucose hypometabolism and
atrophy [21],[27]. However, the previously proposed approaches
only used single modality information for the classification
algorithms loosing this way the differential information which
various biomarkers might provide for a better detection and
differentiation of dementia syndromes.
Here, we apply SVM as the most frequently used multivariate
approach to evaluate its contribution for detection and differentiation of dementia in multimodal imaging. To increase the
validity of our method we apply SVM classification on data
extracted from ROIs based on disorder-specific metabolic
reductions and atrophy reported in comprehensive meta-analyses
investigating AD and FTLD. This method allows a better
generalization of our classification algorithms to other clinical
centers and ensures that only disorder-specific changes are used for
SVM based discrimination. We hypothesize that common use of
different imaging modalities might substantially improve early
detection and differentiation of dementia.
Table 1. Subject group characteristics.
Controls
ANOVA
(df,F,P)
13
21
14
Male/Female
7/6
9/12
7/7
Age (years)
53.966.0
61.166.7
60.866.4
2, 5.76, 0.006
CDR (score)
0.2360.26
0.7160.25
0.8260.42
2, 13.93, 0.000
MMSE (score)
n.a.
23.263.9
24.464.2
10.763.1
11.663.8
2,1.02,0.368
Mean 6 standard deviation. AD Alzheimers disease, ANOVA analysis of

variance, CDR Clinical Dementia Rating Scale, FTLD frontotemporal lobar
degeneration, MMSE Mini Mental State Examination, n.a. not available.
lesions due to stroke, traumatic head injury, brain tumor or

inflammatory diseases.
Data acquisition
MRI data. For each subject, a high-resolution T1-weighted
MRI scan was obtained, consisting of 128 sagittal slices adjusted to
AC-PC line and a with slice thickness of 1.5 mm and pixel size of
161 mm2. MRI was performed on two different 3T scanners
(MedSpec 30/100, Bruker Biospin, Ettlingen Germany and
Magnetom Trio, Siemens, Erlangen, Germany) using two
different T1-weighted sequences (MDEFT or MP-RAGE with
TR = 1300 ms, TI = 650 ms, TE = 3.93 ms or TE = 10 ms; FOV
25625 cm2; matrix = 2566256 voxels). On the MedSpec
scanner, only the MDEFT-sequence and on the Magnetom Trio
scanner, either MDEFT or MP-RAGE sequences were used. The
distribution of scanner types and sequences used to obtain the
MRI data was random across subjects and did not differ
significantly in its distribution between the groups nor for
scanner type nor for sequence.
PET data. Each subject also underwent FDG-PET imaging
either a few a weeks before or after the MRI scan. All PET data
were acquired on a Siemens ECAT EXACT HR+ scanner (CTI/
Siemens, Knoxville, TN, USA) under a standard resting condition
in 2-dimensional (2D) mode. The 2D acquisition mode was used
because it allows a better quantification of the PET data due to
lower scatter radiation. Sixty-three slices were simultaneously
collected with an axial resolution of 5 mm full width at half
maximum (FWHM) and in-plane resolution of 4.6 mm. After
correction for attenuation, scatter, decay and scanner-specific
dead-time, images were reconstructed by filtered back-projection
using a Hann-filter of 4.9 mm FWHM. The 63 transaxial slices
obtained had a matrix of 1286128 voxels with an edge length of
2.45 mm.
Ethics Statement
The research protocol was approved by the Ethics Committee
of the University of Leipzig, and was in accordance with the latest
version of the Declaration of Helsinki. Informed consent was
obtained from all subjects.
Subjects
We analyzed FDG-PET and T1-weighted MRI data of 21
patients (Table 1) with an early stage of probable AD, 14 patients
with an early stage of FTLD and 13 control subjects. Patients were
recruited from the Day Clinic of Cognitive Neurology at the
University of Leipzig. Probable AD was diagnosed according to
NINCDS-ADRDA criteria [28]. Although all AD subjects also
fulfilled the revised NINCDS-ADRDA criteria suggested by
Dubois et al. [17] the fulfillment of the original McKhann criteria
was sufficient for the inclusion into the study. Diagnosis of FTLD
was based on criteria suggested by Neary et al. [29]. The control
group included subjects who visited the Day Clinic with subjective
cognitive complaints, which were not objectively confirmed by a
comprehensive neuropsychological and clinical evaluation. FDGPET and MRI for these subjects was conducted for diagnostic
reasons within the clinical assessment. This control group was
chosen because, in clinical practice, it is crucial to discriminate
between these subjects showing a normal age-related decrease in
cognitive performance and patients with an early stage of
dementia. Patients were excluded if structural imaging revealed
FTLD
Number
Education (years) 12.363.1
Methods
90
AD
Image processing and statistical analysis

The procedure described below has been specifically designed
for this study, aiming at a most accurate co-processing of FDGPET and MRI data to obtain a more precise between subject
anatomical overlap (Figure 1). All image-processing steps were
carried out using the SPM5 software package (Statistical
Parametric Mapping software: http://www.fil.ion.ucl.ac.uk/
spm/) implemented in Matlab 7.7 (MathWorks Inc., Sherborn,
MA). SVM classification was conducted with the LIBSVM
software [30] using the Matlab interface.
MR images. The MR images were first interpolated to get an
isotropic resolution of 16161 mm3. The resultant MR images
2

Figure 1. Schematic representation of the procedure for FDG-PET and MRI data handling and processing steps. FDG-PET
[F18]fluorodeoxyglucose positron emission tomography, MRI magnetic resonance imaging, MNI Montreal Neurological Institute, PVE partial volume
effect correction.
were coregistered on their respective FDG-PET images and bias

corrected for inhomogeneity artifacts using the Unified
Segmentation Approach described in detail by Ashburner and
Friston [31]. This specific method performs a better coregistration
of images from different modalities and allows a more accurate
segmentation due to the bias correction. A further reason to use
this approach was that the straightforward coregistration
implemented in the PVElab software described later sometimes
failed. We used this software for automatic partial volume
correction of the FDG-PET images. The coregistered MR
images were processed using the DARTEL (Diffeomorphic
Anatomical Registration Through Exponentiated Lie algebra)
approach [32] to enable a more accurate spatial normalization.
This approach registers all gray matter (GM) and white matter
(WM) images to an averaged-size template created from all
subjects used in this study and modulates the images to preserve
the total amount of signal from each region in the images.
Subsequently, the images were smoothed using a Gaussian kernel
of 12 mm FWHM. This smoothing factor, although higher then
usual MR kernels, was selected based on extensive tests, because it
allows the optimal coevaluation with lower resolution FDG-PET
images.
FDG-PET data. Within the common registration with MRI
data using the Unified Segmentation Approach described above,
PET images were interpolated to the same voxel size as the MR
images, namely 16161 mm3. This processing does not introduce
any additional noise into the PET images. However, in our
experience, it substantially improves the subsequent partial volume
effect (PVE) correction of voxels representing GM intensities using
the modified Muller-Gartner method [33],[34]. Due to the

interpolation, they are exactly overlaid with the MR tissue class
images of the same subject obtained from the segmentation step in
the PVE approach. Thus, the within-voxel correction is done only
for those voxels directly overlaying the GM structures in the MR
images. Instead of smoothing the MR data to the resolution of
PET data and thus loosing the exact quantitative and qualitative
information of GM distribution, which is usually done in the PVE
correction, the interpolation of FDG-PET preserves this
information. This allows a more accurate correction of atrophy
effects onto glucose utilization. The subsequent PVE correction
including all image processing steps was done by using the
automatic algorithm implemented in the PVElab software package
[35]. Because the modified Muller-Gartner method sets all WM
voxel values to the mean WM intensity value, these regions do not
contain any further valuable regional information after the PVE
correction. For this reason, all voxels belonging to WM were
masked using the ImCalc function in the SPM5 software package
by filtering this specific intensity in the whole image. After the PVE
correction, the DARTEL flow fields calculated from the MR
images were applied to their respective PET images to obtain an
anatomically exact overlap between GM and PET images of all
subjects with modulation to preserve the total amount of signal
from each region. In the same way as the MR data, the PET data
were smoothed by a Gaussian kernel of 12 mm FWHM. Finally,
the FDG-PET data were intensity normalized using cerebellar
ROIs to account for individual differences in global PET
measures. This region has been shown to be least affected in
mild to moderate stages of AD [36]. Additionally, normalization to
3
91
92
this region improves the statistical discrimination between

dementia patients and control subjects in comparison to other
regions reported in the literature [37][39].
Masking. The MR and PET images obtained as described
above were masked to avoid contamination by misclassified voxels.
Voxels lying between WM and ventricular cerebrospinal fluid tend
to be misclassified as GM voxels due to their similar intensity. The
mask was obtained after extensive testing by excluding all voxels in
the first and the last template created by the DARTEL approach
with a probability of below 0.2 for belonging to GM and including
only the voxels that exceed this threshold in both templates. This
mask was applied twice: firstly prior to smoothing to avoid
misclassification, and secondly, after the smoothing to avoid big
edge effects. WM images were exclusively masked using the same
mask to avoid overlaps between GM and WM voxels due to
smoothing. The masked images were used for the subsequent
SVM analysis of the data.
ROI extraction. ROI coordinates (Table 2) were extracted
from two comprehensive, systematic and quantitative metaanalyses investigating biomarkers of AD and FTLD in MR and
FDG-PET images. The meta-analyses included a total number of
1618 patients (AD/FTLD: 1351/267) and 1448 healthy control
subjects (1097/351) [8],[9]. These meta-analyses extracted the
prototypical networks of AD and FTLD by applying what is
currently the most sophisticated and best-validated of coordinatebased voxel-wise meta-analyses, anatomical likelihood estimate. In
the FTLD meta-analyses [8] only coordinates which are common
to all subgroups of FTLD patients were used. In total, 10 regions
from MRI and 6 regions from FDG-PET were used from the
FTLD meta-analysis. From AD meta-analysis, 6 regions were used
from both, MRI and FDG-PET. The AD meta-analysis also
identified one additional region in the fornix which was
differentiating between early-onset AD patients (age,65 years)
and control subjects but not between late-onset AD patients and
control subjects. This region was not included into the ROI-based
classification to avoid a discrimination bias towards early-onset
AD patients. Although unequal numbers of ROIs were used from
both imaging modalities, this number is also a highly important
information as it also provides a measurement for the amount of
changes present in a specific modality. Because the coordinates in
both meta-analyses were reported in the Talairach space, they
were transformed to MNI space according to a formula proposed
by Matthew Brett (published on the Internet: http://www.mrccbu.
cam.ac.uk/Imaging/Common/mnispace.shtml). DARTEL preprocessed data are registered to an averaged size template created from
all subjects in this study. To transform these data to the MNI space we
normalized them to an a priori MNI template in SPM by using affineonly spatial normalization. Due to the affine-only transformation, our
images still differed in shape from the MNI template, so some reported
coordinates were slightly outside of the anatomic regions in our
imaging data. In this case, the center coordinates for the ROIs were
moved slightly towards the closest point of the corresponding
anatomical region reported in the meta-analyses. ROIs were selected
using the 3D fill tool in the MRIcron software package (http://www.
sph.sc.edu/comd/rorden/mricron). Separate ROI masks were created
for MR and FDG-PET images based on the origin of the peak values
reported in the meta-analyses and using all regions reported for AD
and FTLD in a single mask for each modality. Each ROI was
restricted to a sphere with a radius of 5 mm around the reported
coordinate (Figure 2). Additionally, to increase the signal-to-noise ratio,
all zero voxels and edge voxels with an intensity deviation of 13
intensity units in the MRIcron 3D fill tool were excluded from the
ROI. The edge voxel restriction excludes all voxels at the edge of the
smoothed GM structures within the sphere. These voxels carry much
Table 2. Coordinates of ROIs used for SVM classification.
FTLD vs. Controls

FDG-PET
BA
Lat x
Pregenual anterior cingulate gyrus
24/32
25
Lentiform nucleus; Caudate head
220
Medial thalamus
22
219 6
34 21
23
Anterior insula
15/16
247
10 27
Anterior medial frontal cortex
10
54 0
25
22 225
Amygdala
MRI
BA
Lat x
25
49 25
24/32
25
34 21
Inferior frontal gyrus, pars opercularis
44
254
11 20
220
247
10 27
Lentiform nucleus; Caudate head

Anterior insula
15/16
Subcallosal/septal area
25
23
22
10 210
Amygdala
25
22 225
Amygdala; Ento-and perirhinal cortex
226
25 225
Temporal pole
38
229
11 243
Temporal pole
38
247
244
BA
Lat x
AD vs. Controls
FDG-PET
Angular gyrus
39
238
268 37
Angular gyrus
39
43
268 33
Posterior superior temporal sulcus
21/22
R
31 31
236 27
9/10
32
Inferior precuneus
31
Dorsal posterior cingulate cortex
23
Posterior superior temporal sulcus
21/22
251
261 23
Middle inferior temporal sulcus
20/21
59
231 223
MRI
BA
Lat x
Posterior insula
13
238
225 15
Medial thalamus
25
213 3
Hippocampal body/tail
31
238 26
21/22
263
221 25
Amygdala, anterior hippocampal formation, 28/34

uncus, (trans-) entorhinal area
25
28 218
Amygdala, anterior hippocampal formation, 28/34

uncus, (trans-) entorhinal area
226
28 218
Middle temporal gyrus/superior temporal

sulcus
Coordinates are in MNI space (L left, R right). AD Alzheimers disease, BA

Brodmann area, FDG-PET [18F]fluorodeoxyglucose positron emission
tomography, MRI magnetic resonance imaging, ROI region-of-interest, SVM
support vector machine.
less information due to their further distance from the GM structures in

the unsmoothed data and so decrease the signal-to-noise ratio in the
corresponding ROI.
SVM. Multivariate pattern classification, as described in
Kloppel et al. [10], was performed with a linear kernel by
identifying a separating hyperplane that maximizes the distance
4

classification is a highly memory-consuming approach. To ensure

that our classification results were not based on factors randomly
discriminating between groups, we reran the whole-brain and
ROI-based classification for comparison 30 times by randomly
assigning all subjects to the three groups independently from the
clinical diagnosis and calculating the classification accuracy by
using the leave-one-out procedure described above.
Statistical analysis. Group comparisons for age, education
and CDR (Clinical Dementia Rating Scale) [40] were performed
by conducting ANOVAs (analyses of variance). If an ANOVA
revealed a significant between-group effect, a Bonferroni t-test was
calculated with a significance threshold of p,05 (corrected for
multiple comparisons, two-tailed). MMSE (Mini Mental State
Examination) [41], was only present for 20 patients with AD and
11 patients with FTLD. MMSE scores of these two groups were
compared to each other using an independent samples t-test.
Group differences regarding sex were evaluated using a chi-square
test for independent samples. The statistical analysis was
performed with the commercial software package SPSS 17.0
(http://www.spss.com/statistics/).
Figure 2. Regions of interest extracted from gray matter (left)

and FDG-PET (right) data for AD and FTLD patients and used
for support vector machine classification projected onto a
glass brain (top) and onto an axial slice (bottom). AD Alzheimers
disease, FDG-PET [F18]fluorodeoxyglucose positron emission tomography, FTLD frontotemporal lobar degeneration, GM gray matter, ROIs
regions of interest.
Results
Clinical characteristics
The chi-square test for independent samples did not reveal any
statistical differences in sex between the groups [x2(2) = 0.42;
p = 0.809]. The three groups did not differ in education (Table 1).
CDR scores differed significantly in the three groups. The post-hoc
test revealed no differences in the mean CDR scores between both
groups of dementia patients [t(33) = 0.94;p = 0.977]. As expected,
both early AD [t(32) = 5.36;p,0.001] and early FTLD
[t(25) = 4.35;p,0.001] had significantly higher CDR scores compared to the control subjects. MMSE scores also did not differ
between both groups of dementia patients indicating a similar severity
of dementia syndrome [t(29) = 281;p = 0.95]. The ANOVA also
revealed a significant group difference in age. The two groups of
dementia patients did not differ significantly in age
[t(33) = 0.16;p = 1.0]. There was a minor but significant difference
between AD patients and controls [t(32) = 3.18;p = 008] and FTLD
patients and controls [t(25) = 2.86; p = 024].
between different clinical groups based on whole-brain or ROIs

information. The cross-validation of the trained SVM was
performed by using the leave-one-out method. This procedure
iteratively leaves out the information of each subject and trains the
model on the remaining subjects for subsequent class assignation
of the person that was not included in the training procedure. This
validation method enables the generalization of the trained SVM
to data that have never been presented to the SVM algorithms
previously. The reported accuracy is the percentage of subjects
correctly assigned to the clinical diagnosis. Usually SVM
classification is performed without smoothing of the data,
because single voxels are assumed to contain information, for
example, for prediction of future action based on functional MR
images. However, in neurodegenerative disorders single voxels are
unlikely to contain generalizable information due to a limited
across-subject registration of MR and FDG-PET images.
Although SVM classification based on unsmoothed data has
been shown to differentiate reasonably between different groups
(Kloppel et al., 2007), an additional smoothing should make this
approach more reliable and generalizable to new data. To control
for the effect of smoothing we ran the same whole-brain
classification twice for GM, PET and for integration of GM and
PET in the same vector with and without smoothing.
We performed the whole-brain SVM classification using GM,
WM or FDG-PET images separately and by combining
information from different modalities. For the SVM classification,
all data of a subject are transformed into a vector, with
information of an additional modality simply attached by
extending the vector. Additionally, we repeated the whole-brain
SVM classification by adding MR to FDG-PET information
combining both modalities in a single image. ROI-based SVM
classification was performed on data extracted from smoothed
images separately for GM and FDG-PET images and also by
integrating information from both modalities in a single vector. In
order to reduce the number of voxels in the ROI-based
classification, only nonzero voxels were included in the vector.
This was done because otherwise the whole-brain SVM
SVM Whole-brain analysis

Multivariate classification of the data using SVM at the wholebrain level revealed the best discrimination accuracy for all three
groups using FDG-PET, with 81% (chance level 33%), in comparison
to GM and WM information, with lowest accuracy using WM
information on its own (Table 3). The combination of metabolism
and GM values in a single image revealed a similar accuracy for
differentiation of the three groups, with higher accuracy for
differentiation between both types of dementia, however, with
slightly lower discrimination between dementia patients and control
subjects. Whole-brain SVM classification for the three groups without
smoothing revealed lower accuracy rates in all classifications in
comparison to differentiation based on smoothed images. The
accuracy increase due to smoothing ranged between 2 (GM) and 6%
(FDG-PET). Figure 3 displays regions that were most influential in
making binary classification between the AD, FTLD and control
subjects based on smoothed whole-brain information.
SVM ROI analysis

Accuracy based on ROIs from both meta-analyses using only
GM information was substantially lower for differentiation
between AD and FTLD patients in comparison with whole-brain
classification. However, it was comparable to the whole-brain
5
93
Table 3. Accuracy rates for whole-brain and ROI-based SVM classification for FDG-PET and MRI.
AD, FTLD and Controls
AD vs FTLD
AD vs Controls
FTLD vs Controls
GM
whole-brain
72.9%
80.0%
88.2%
77.8%
WM
whole-brain
66.7%
74.3%
79.4%
77.8%
FDG-PET
whole-brain
81.3%
82.9%
94.1%
92.6%
GM/ FDG-PET
whole-brain
79.2%
82.9%
94.1%
88.9%
GM/WM/FDG-PET
whole-brain
77.1%
82.9%
91.2%
85.2%
GM + FDG-PET
whole-brain
81.3%
88.6%
91.2%
88.9%
GM
ROIs
56.3%
60.0%
82.4%
85.2%
FDG-PET
ROIs
75.0%
80.0%
94.1%
85.2%
GM/FDG-PET
ROIs
91.7%
94.3%
100.0%
92.6%
Accuracy represents the percentage of subjects correctly assigned to the correct condition. AD Alzheimers disease, FDG-PET [18F]fluorodeoxyglucose positron emission
tomography, FTLD frontotemporal lobar degeneration, GM gray matter, MRI magnetic resonance imaging, ROI region-of-interest, SVM support vector machine, WM
white matter.
approach in differentiating between patients with both types of

dementia and control subjects. ROIs extracted from FDG-PET
data showed slightly lower discrimination accuracy compared to
whole-brain information. The best accuracy rates of all SVM
classifications were obtained using combined information extracted from FDG-PET and GM data. This approach resulted in a
classification accuracy of 92% for the differentiation of all three
groups and an accuracy rate of 94% for differentiation between
AD and FTLD patients. Sensitivity of this ROI-based classification
ranged between 85.7% for FTLD and 100% for AD and
specificity of 100% for discrimination of both types of dementia
from control subjects (Table 4).
The classification accuracy in the 30 trials randomly assigning
all subjects to the three groups resulted in a mean accuracy rate of
3467.7% (mean 6 standard deviation), ranging between 21 and
52% for the ROI-based SVM classification, and 33.768.2%
ranging between 12 and 50% for the whole-brain classification.
by previous studies [10],[12],[13],[24]. It not only captures

univariate relationships of a single voxel across all subjects but is
also able to detect multivariate relationships over a large group of
information, as, for example, between different structures and
modalities in the brain. Furthermore, this tool provides an easy
way to use this information for classifying imaging data of new
subjects to a specific condition.
Here, we systematically compared different information provided by FDG-PET and MRI to enable the most accurate
detection and differentiation of dementia. The diagnosis was based
on comprehensive clinical and neuropsychological testing. Although the data are not histopathologically confirmed to be sure of
assigning them to the correct condition, generally higher
conformity with the clinical diagnosis should also result in more
accurate classification of histopathologically validated data.
The whole-brain SVM classification provided the most accurate
classification using only FDG-PET information. GM and WM
based classification accuracy was lower for all comparisons
indicating a lower sensitivity for detection of dementia-relevant
information. Nonetheless, classification based on GM, WM and
FDG-PET separately or combining them revealed a discrimination accuracy which was above chance level for the correct
categorization of the three groups. All classification results
substantially exceeded the best classification accuracy obtained
by randomly assigning all subjects to different groups. Additionally, smoothing of the data improved the classification accuracy in
both imaging modalities as expected.
However, in whole-brain classification noise is introduced by
using a great deal of information for classification that does not
differentiate between the groups. Recent comprehensive metaanalyses identified the prototypical networks for both disorders
in both modalities using VBM [8],[9]. The involved regions have
been shown to be affected in AD and FTLD patients most
consistently in all studies investigating these disorders. By using this
information, we ruled out the possibility that our classification
results are dependent to our group of patients. Although this
Discussion
In this study we performed a multimodal comparison and
discrimination of dementia patients using FDG-PET and MRI. To
enable a more accurate coevaluation of both imaging modalities,
we developed a new preprocessing algorithm. This algorithm was
designed to enable an accurate anatomical registration of both
modalities. All processing steps were performed as far as possible
simultaneously by applying the same deformations and preprocessing parameters to both modalities of the same subject. This
procedure resulted in an accurate anatomical overlap of both
imaging modalities and in an accurate between-subject registration, with both images having the same voxel size and
approximately the same effective smoothness.
SVM
94
SVM classification is a very promising tool for detection and

differentiation of different dementia syndromes, as has been shown

whole-brain classification with the highest accuracy gain for the

differentiation of both types of dementia, which, with 94%, is the
highest differentiation rate reported up to now. Accordingly, we
suggest this method as a diagnostic standard for the classification
of dementia syndromes.
Nonetheless, some limitations should be considered regarding
the results of the present study. First of all, the number of subjects
used for classification is too low to allow a generalizable conclusion
for patients from other clinical centers. This is especially a problem
because the accuracy of the clinical diagnosis for specific dementia
syndromes, which was used here for validation, strongly varies
between different clinical centers. Therefore, in a future work this
approach should be validated using a larger and more generalizable dataset like the data provided by the Alzheimers disease
Neuroimaging Initiative (ADNI: www.adni-info.org). A further
limitation of the present work is the significantly younger age of
the control group in comparison to the patient cohort. However,
this aspect might only have contributed to the discrimination
between dementia patients and control subjects but not to the high
classification accuracy of AD and FTLD patients as these were
very similar in their age range. If age contributed to the
classification accuracy there should be lower classification
accuracy for young dementia patients and older control subjects
as they did not differ in age. For comparison of both types of
dementia patients and control subjects the classification accuracy
did not differ for younger and older dementia patients although
half of the patients were in the same age range as the control
group. In AD group all patients were classified correctly. In FTLD
group one younger and one older patient were misclassified.
Independently of age, all control subjects were classified correctly
for both comparisons. These results indicate that the slight mean
age differences is not the decisive factor for the high discrimination
accuracy using combined information from FDG-PET and MRI.
Furthermore, if age still slightly contributed to the high
discrimination of dementia patients and control subjects this
contribution was also present in all other single modality and
multimodal whole-brain and ROI-based SVM classifications
applied in this study. Therefore, age cannot account for increased
differentiation accuracies when combined ROI information from
FDG-PET and MRI are used for differentiation of dementia
patients and control subjects.
Another point is that subjects in the control group in our study
reported subjective cognitive complaints which might have limited
the interpretation of the results of our study. However, only
subjects were included whose cognitive complaints were not
confirmed by comprehensive neuropsychological evaluation. The
CDR is a semi-structural interview and is highly dependent on the
subjectively perceived memory impairment which resulted in a
CDR score of 0.5 for these control subjects in our study. In recent
literature it has been shown that the CDR stage of 0.5 has a poor
discriminative value for healthy control subjects and subjects with
Mild Cognitive Impairment (MCI) [42],[43]. Meguro et al. [42]
have shown that about 30% of a normal population older then 65
got a CDR score of 0.5 while the prevalence of MCI in the same
population was only about 5% which suggests that CDR score of
0.5 is not a good indicator of MCI. Due to the absence of any
objective cognitive impairment in all neuropsychological tests for
all subjects included in the control group in our study this group of
subjects can be regarded as cognitively unimpaired.
Figure 3. Weights of voxels most relevant for classification of

both groups of patients and control subjects in FDG-PET and
MRI after SVM training. Weights are relative, and have no applicable
units. AD and FTLD vs Controls: Blue and light blue indicate decreased
gray matter intensity (upper row) or reduced metabolic rate (lower row)
that increase the likelihood of classification into a dementia group. Red
and yellow indicate the opposite. AD vs FTLD: Blue and light blue
indicate decreased gray matter intensity (upper row) or reduced
metabolic rate (lower row) that increase the likelihood of classification
into the AD group. Red and yellow indicate the opposite. Regions with
bright colors (yellow and light blue) have a higher importance for
separation than regions with dark colors (blue and red). AD Alzheimers
disease, FTLD frontotemporal lobar degeneration. MRI magnetic
resonance imaging, PET positron emission tomography.
method provides lower accuracy rates for GM or FDG-PET

information on their own, it shows a significantly higher
discrimination rate by combining both information modalities
into a single vector. This ROI-based discrimination is superior to
Table 4. Differentiation rates for combined region-of-interest
information from FDG-PET and MRI.
Accuracy
Sensitivity
Specificity
92.9%
AD vs FTLD
94.3%
95.2%*
AD vs Controls
100.0%
100.0%
100.0%
FTLD vs Controls
92.6%
85.7%
100.0%
Conclusion and perspectives
*Considering a correctly identified AD as a true positive. AD Alzheimers disease,

FDG-PET [18F]fluorodeoxyglucose positron emission tomography, FTLD
frontotemporal lobar degeneration, MRI magnetic resonance imaging.
In our study, we investigated the advantages of SVM

classification using combined information from FDG-PET and
MRI to improve detection and differentiation of dementia.
7
95
Furthermore, based on affected regions reported in previous

studies, investigating Alzheimers disease and frontotemporal lobar
degeneration with univariate approaches and summarized in two
meta-analyses, we applied linear support vector machine classification algorithm using information from both imaging modalities.
Combining region-of-interest information from FDG-PET and
MRI resulted in a substantial gain in accuracy compared to wholebrain and to single modality classification for both detection and
differentiation of Alzheimers disease and frontotemporal lobar
degeneration. Our results indicate that integration and combina-
tion of results from different imaging modalities might provide a

new way to improve the diagnostic accuracy of these dementia
disorders.
Conceived and designed the experiments: JD KM AH HB HEM AV OS
MLS. Performed the experiments: JD KM AH HB HEM AV OS MLS.
Analyzed the data: JD KM AH HB HEM AV OS MLS. Contributed
reagents/materials/analysis tools: JD KM AH HB HEM AV OS MLS.
Wrote the paper: JD KM AH HB HEM AV OS MLS.
References
1. Hoffman JM, Welsh-Bohmer KA, Hanson M, Crain B, Hulette C, et al. (2000)
FDG PET imaging in patients with pathologically verified dementia. J Nucl Med
41: 19201928.
2. Rosen HJ, Gorno-Tempini ML, Goldman WP, Perry RJ, Schuff N, et al. (2002)
Patterns of brain atrophy in frontotemporal dementia and semantic dementia.
3. Diehl J, Grimmer T, Drzezga A, Riemenschneider M, Forstl H, et al. (2004)
Cerebral metabolic patterns at early stages of frontotemporal dementia and
semantic dementia. A PET study. Neurobiol Aging 25: 10511056.
4. Jeong Y, Cho SS, Park JM, Kang SJ, Lee JS, et al. (2005) 18F-FDG PET
findings in frontotemporal dementia: an SPM analysis of 29 patients. J Nucl Med
46: 233239.
5. Diehl-Schmid J, Grimmer T, Drzezga A, Bornschein S, Riemenschneider M,
et al. (2007) Decline of cerebral glucose metabolism in frontotemporal dementia:
a longitudinal 18F-FDG-PET-study. Neurobiol Aging 28: 4250.
6. Fung G, Stoeckel J (2007) SVM feature selection for classification of SPECT
images of Alzheimers disease using spatial information. Knowledge and
Information Systems 11: 243258.
7. Sabri O, Kendziorra K, Wolf H, Gertz HJ, Brust P (2008) Acetylcholine
receptors in dementia and mild cognitive impairment. Eur J Nucl Med Mol
Imaging 35(Suppl 1): S3045.
8. Schroeter ML, Raczka K, Neumann J, Yves von Cramon D (2007) Towards a
nosology for frontotemporal lobar degenerations-a meta-analysis involving 267
subjects. Neuroimage 36: 497510.
9. Schroeter ML, Stein T, Maslowski N, Neumann J (2009) Neural correlates of
Alzheimers disease and mild cognitive impairment: a systematic and
quantitative meta-analysis involving 1351 patients. Neuroimage 47: 11961206.
Automatic classification of MR scans in Alzheimers disease. Brain 131:
681689.
11. Matsunari I, Samuraki M, Chen WP, Yanase D, Takeda N, et al. (2007)
Comparison of 18F-FDG PET and optimized voxel-based morphometry for
detection of Alzheimers disease: aging effect on diagnostic performance. J Nucl
Med 48: 19611970.
biomarkers of prodromal Alzheimers disease: a high-dimensional pattern
classification study. Neuroimage 41: 277285.
14. Sadeghi N, Foster NL, Wang AY, Minoshima S, Lieberman AP, et al. (2008)
Automatic classification of Alzheimers disease vs. Frontotemporal dementia: A
spatial decision tree approach with FDG-PET. 2008 Ieee International
Symposium on Biomedical Imaging: From Nano to Macro Vols 1-4: 408411.
15. Knopman DS, Boeve BF, Parisi JE, Dickson DW, Smith GE, et al. (2005)
Antemortem diagnosis of frontotemporal lobar degeneration. Ann Neurol 57:
480488.
16. Rascovsky K, Salmon DP, Ho GJ, Galasko D, Peavy GM, et al. (2002) Cognitive
profiles differ in autopsy-confirmed frontotemporal dementia and AD.
Neurology 58: 18011808.
17. Dubois B, Feldman HH, Jacova C, Dekosky ST, Barberger-Gateau P, et al.
NINCDS-ADRDA criteria. Lancet Neurol 6: 734746.
18. Kipps CM, Hodges JR, Fryer TD, Nestor PJ (2009) Combined magnetic
resonance imaging and positron emission tomography brain imaging in
behavioural variant frontotemporal degeneration: refining the clinical phenotype. Brain 132: 25662578.
19. Ishii K, Willoch F, Minoshima S, Drzezga A, Ficaro EP, et al. (2001) Statistical
brain mapping of 18F-FDG PET in Alzheimers disease: validation of anatomic
standardization for atrophied brains. J Nucl Med 42: 548557.
20. Ishii K, Sasaki H, Kono AK, Miyamoto N, Fukuda T, et al. (2005) Comparison
of gray matter and metabolic reduction in mild Alzheimers disease using FDGPET and voxel-based morphometric MR studies. Eur J Nucl Med Mol Imaging
32: 959963.
21. Kanda T, Ishii K, Uemura T, Miyamoto N, Yoshikawa T, et al. (2008)
Comparison of grey matter and metabolic reductions in frontotemporal
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
dementia using FDG-PET and voxel-based morphometric MR studies.

Eur J Nucl Med Mol Imaging 35: 22272234.
van de Pol LA, Hensel A, van der Flier WM, Visser PJ, Pijnenburg YA, et al.
(2006) Hippocampal atrophy on MRI in frontotemporal lobar degeneration and
Alzheimers disease. J Neurol Neurosurg Psychiatry 77: 439442.
Schroeter ML, Raczka K, Neumann J, von Cramon DY (2008) Neural networks
in frontotemporal dementia a meta-analysis. Neurobiol Aging 29: 418426.
Chaves R, Ramirez J, Gorriz JM, Lopez M, Salas-Gonzalez D, et al. (2009) SVMbased computer-aided diagnosis of the Alzheimers disease using t-test NMSE
feature selection with feature correlation weighting. Neurosci Lett 461: 293297.
Horn JF, Habert MO, Kas A, Malek Z, Maksud P, et al. (2009) Differential
automatic diagnosis between Alzheimers disease and frontotemporal dementia
based on perfusion SPECT images. Artif Intell Med 47: 147158.
Ramirez J, Gorriz JM, Salas-Gonzales D, Romero A, Lopez M, et al. (2009)
Computer-aided diagnosis of Alzheimers type dementia combining support
vector machines and discriminant set of features. Information Sciences:
doi:10.1016/j.ins.2009.05.012.
Rabinovici GD, Seeley WW, Kim EJ, Gorno-Tempini ML, Rascovsky K, et al.
(2007) Distinct MRI atrophy patterns in autopsy-proven Alzheimers disease and
frontotemporal lobar degeneration. Am J Alzheimers Dis Other Demen 22: 474488.
McKhann G, Drachman D, Folstein M, Katzman R, Price D, et al. (1984)
Neary D, Snowden JS, Gustafson L, Passant U, Stuss D, et al. (1998)
Frontotemporal lobar degeneration: a consensus on clinical diagnostic criteria.
Neurology 51: 15461554.
Chang CC, Lin CJ (2001) LIBSVM: a library for support vector machines.
Software available at: http://www.csie.ntu.edu.tw/ cjlin/libsvm.
Ashburner J, Friston KJ (2005) Unified segmentation. Neuroimage 26: 839851.
Ashburner J (2007) A fast diffeomorphic image registration algorithm. Neuroimage 38: 95113.
Muller-Gartner HW, Links JM, Prince JL, Bryan RN, McVeigh E, et al. (1992)
Measurement of radiotracer concentration in brain gray matter using positron
emission tomography: MRI-based correction for partial volume effects. J Cereb
Blood Flow Metab 12: 571583.
Rousset OG, Ma Y, Evans AC (1998) Correction for partial volume effects in
PET: principle and validation. J Nucl Med 39: 904911.
Quarantelli M, Berkouk K, Prinster A, Landeau B, Svarer C, et al. (2004)
Integrated software for the analysis of brain PET/SPECT studies with partialvolume-effect correction. J Nucl Med 45: 192201.
Ishii K, Sasaki M, Kitagaki H, Yamaji S, Sakamoto S, et al. (1997) Reduction of
cerebellar glucose metabolism in advanced Alzheimers disease. J Nucl Med 38:
925928.
Yakushev I, Landvogt C, Buchholz HG, Fellgiebel A, Hammers A, et al. (2008)
Choice of reference area in studies of Alzheimers disease using positron emission
tomography with fluorodeoxyglucose-F18. Psychiatry Res 164: 143153.
Yakushev I, Hammers A, Fellgiebel A, Schmidtmann I, Scheurich A, et al.
(2009) SPM-based count normalization provides excellent discrimination of mild
Alzheimers disease and amnestic mild cognitive impairment from healthy aging.
Dukart J, Mueller K, Horstmann A, Vogt B, Frisch S, et al. (2010) Differential
effects of global and cerebellar normalization on detection and differentiation of
dementia in FDG-PET studies. Neuroimage 49: 14901495.
Morris JC (1993) The Clinical Dementia Rating (CDR): current version and
Folstein MF, Folstein SE, McHugh PR (1975) Mini-Mental State (a practical
method for grading the state of patients for the clinician). Journal of Psychiatric
Research 12: 189198.
Meguro K, Ishii H, Yamaguchi S, Ishizaki J, Sato M, et al. (2004) Prevalence
and cognitive performances of clinical dementia rating 0.5 and mild cognitive
impairment in Japan. The Tajiri project. Alzheimer disease and associated
disorders 18: 310.
Perneczky R, Wagenpfeil S, Komossa K, Grimmer T, Diehl J, et al. (2006)
Mapping scores onto stages: mini-mental state examination and clinical
dementia rating. The American journal of geriatric psychiatry: official journal
of the American Association for Geriatric Psychiatry 14: 139144.
96
Plasma Based Markers of [11C] PiB-PET Brain Amyloid

Burden
Steven John Kiddle1,2*, Madhav Thambisetty3, Andrew Simmons1,2, Joanna Riddoch-Contreras2,
Abdul Hye2, Eric Westman1,2, Ian Pike4, Malcolm Ward4, Caroline Johnston1,2, Michelle
Katharine Lupton2, Katie Lunnon2, Hilkka Soininen5, Iwona Kloszewska6, Magda Tsolaki7, Bruno Vellas8,
Patrizia Mecocci9, Simon Lovestone1,2, Stephen Newhouse1,2., Richard Dobson1,2.*, for the Alzheimers
Disease Neuroimaging Initiative
1 National Institute of Health Research Biomedical Research Centre for Mental Health, South London and Maudsley National Health Service Foundation Trust, London,
United Kingdom, 2 Kings College London, Institute of Psychiatry, London, United Kingdom, 3 Laboratory of Behavioral Neuroscience, National Institute on Aging,
Baltimore, Maryland, United States of America, 4 Proteome Sciences plc, Cobham, Surrey, United Kingdom, 5 School of Neurology, University of Eastern Finland and
University Hospital of Kuopio, Kuopio, Finland, 6 Department of Old Age Psychiatry and Psychotic Disorders, Medical University of Lodz, Lodz, Poland, 7 3rd Department of
Neurology, G. Papanicolaou Hospital, Aristotle University of Thessaloniki, Thessaloniki, Greece, 8 Department of Geriatric Medicine, Grontople de Toulouse, Toulouse
University Hospital, Toulouse, France, 9 Institute of Gerontology and Geriatrics, University of Perugia, Perugia, Italy
Abstract
Changes in brain amyloid burden have been shown to relate to Alzheimers disease pathology, and are believed to precede
the development of cognitive decline. There is thus a need for inexpensive and non-invasive screening methods that are
able to accurately estimate brain amyloid burden as a marker of Alzheimers disease. One potential method would involve
using demographic information and measurements on plasma samples to establish biomarkers of brain amyloid burden; in
this study data from the Alzheimers Disease Neuroimaging Initiative was used to explore this possibility. Sixteen of the
analytes on the Rules Based Medicine Human Discovery Multi-Analyte Profile 1.0 panel were found to associate with [11 C]PiB PET measurements. Some of these markers of brain amyloid burden were also found to associate with other AD related
phenotypes. Thirteen of these markers of brain amyloid burden c-peptide, fibrinogen, alpha-1-antitrypsin, pancreatic
polypeptide, complement C3, vitronectin, cortisol, AXL receptor kinase, interleukin-3, interleukin-13, matrix metalloproteinase-9 total, apolipoprotein E and immunoglobulin E were used along with co-variates in multiple linear regression, and
were shown by cross-validation to explain w30% of the variance of brain amyloid burden. When a threshold was used to
classify subjects as PiB positive, the regression model was found to predict actual PiB positive individuals with a sensitivity of
0.918 and a specificity of 0.545. The number of APOE E 4 alleles and plasma apolipoprotein E level were found to contribute
most to this model, and the relationship between these variables and brain amyloid burden was explored.
Citation: Kiddle SJ, Thambisetty M, Simmons A, Riddoch-Contreras J, Hye A, et al. (2012) Plasma Based Markers of [11C] PiB-PET Brain Amyloid Burden. PLoS
ONE 7(9): e44260. doi:10.1371/journal.pone.0044260
Editor: Ashley I. Bush, Mental Health Research Institute of Victoria, Australia
Received April 13, 2012; Accepted July 31, 2012; Published September 24, 2012
This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for
any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.
Funding: Alzheimers Disease Neuroimaging Initiative (ADNI) data collection and sharing for this project was funded by the ADNI (National Institutes of Health Grant
U01 AG024904). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous
contributions from the following: Abbott; Alzheimers Association; Alzheimers Drug Discovery Foundation; Amorfix Life Sciences Ltd.; AstraZeneca; Bayer HealthCare;
BioClinica, Inc.; Biogen Idec Inc.; Bristol-Myers Squibb Company; Eisai Inc.; Elan Pharmaceuticals Inc.; Eli Lilly and Company; F. Hoffmann-La Roche Ltd and its affiliated
company Genentech, Inc.; GE Healthcare; Innogenetics, N.V.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical
Research & Development LLC.; Medpace, Inc.; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Servier; Synarc Inc.; and
Takeda Pharmaceutical Company. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are
facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and
Education, and the study is coordinated by the Alzheimers Disease Cooperative Study at the University of California, San Diego. ADNI data are disseminated by the
Laboratory for Neuro Imaging at the University of California, Los Angeles. This research was also supported by National Institutes of Health grants P30 AG010129 and K01
AG030514. This study was supported by InnoMed (Innovative Medicines in Europe www.imi.europa.eu/) an Integrated Project funded by the European Union of the
Sixth Framework program priority FP6-2004-LIFESCIHEALTH-5, Life Sciences, Genomics and Biotechnology for Health. This study was also supported by funds from the
National Institutes for Health Research Biomedical Research Centre for Mental Health at the South London and Maudsley National Health Service Foundation Trust and
Institute of Psychiatry, Kings College London. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: Intellectual property has been registered on the use of plasma proteins for use as biomarkers for AD by Kings College London and Proteome
Sciences, with SL and M. Thambisetty named as inventors. IP and MW were full-time employees of Proteome Sciences, London, United Kingdom, at the time of their
contribution to the work described in this manuscript. This does not alter the authors adherence to all the PLOS ONE policies on sharing data and materials. Patent Title
Methods and Compositions Relating to Alzheimers Disease Subject Covers utility of around 30 proteins, specifically listing 16 in the Dependent claims for diagnosis of
Alzheimers disease Filing Proteome Sciences with Kings Business United Kingdom Priority GB0421639.6 dated 29/09/2004 PCT Application PCT/GB2005/003756 dated
29/09/2005 Application in Europe, Japan, United States, Australia, and Canada, dated 15/03/2007 to 16/10/2007. In addition Alzheimers Disease Neuroimaging
Initiative recieved funding from: Abbott; Alzheimers Association; Alzheimers Drug Discovery Foundation; Amorfix Life Sciences Ltd.; AstraZeneca; Bayer HealthCare;
BioClinica, Inc.; Biogen Idec Inc.; Bristol-Myers Squibb Company; Eisai Inc.; Elan Pharmaceuticals Inc.; Eli Lilly and Company; F. Hoffmann-La Roche Ltd and its affiliated
company Genentech, Inc.; GE Healthcare; Innogenetics, N.V.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical
Research & Development LLC.; Medpace, Inc.; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Servier; Synarc Inc.; and
Takeda Pharmaceutical Company. This does not alter the authors adherence to all the PLOS ONE policies on sharing data and materials.
* E-mail: [email protected] (SJK); [email protected] (RD)
September 2012 | Volume 7 | Issue 9 | e44260
97
Plasma Based Markers of Brain Amyloid Burden
complement C3 have previously been shown to associate with

brain amyloid burden [14,15], while others are associated with
other diseases. These assays were performed in plasma samples
that were collected from participants in the Alzheimers Disease
Neuroimaging Initiative (ADNI; http://adni.loni.ucle.edu) study
who also underwent [11 C]-PiB PET imaging for quantification of
fibrillar brain amyloid burden. The main aim of this study was to
ask whether concentrations of a panel of plasma proteins and
metabolites might accurately reflect the extent of fibrillar amyloid
in the brain. A secondary aim was to understand the relationship
between the number of APOE E 4 alleles and plasma based markers
of brain amyloid burden.
Introduction
The failure of several clinical trials targeting brain amyloid
deposition in patients with Alzheimers disease (AD) has led to the
suggestion that these agents may be useful if targeted at older
individuals in pre-symptomatic stages of the disease [1,2].
Screening methods that accurately identify at-risk non-demented
older individuals who are most likely to benefit from such
treatments will therefore represent a major advance in our ability
to effectively test these disease-modifying treatments [3]. If clinical
trials of amyloid lowering interventions were successful in the presymptomatic stages of AD, then there would be a desire to identify
non-demented elderly individuals with elevated brain amyloid
burden who could potentially benefit from early intervention.
However identifying suitable individuals poses a great challenge in
terms of feasibility and cost. To date, the two methods that are
most likely to be useful in estimating levels of brain amyloid
burden are in vivo imaging with positron emission tomography
(PET) using radioligands binding to fibrillar amyloid beta (Ab),
such as [11 C] Pittsburgh B compound (PiB), and assays of Ab levels
in cerebrospinal fluid (CSF) [4,5]. However, both these methods
have inherent drawbacks that limit their utility as screening tools,
especially in resource-poor settings. While PET scanning is
expensive and limited to specialised centres, lumbar puncture to
obtain CSF is associated with some patient discomfort and is
unlikely to be used in primary health care centres to routinely
screen large numbers of elderly patients. An inexpensive, noninvasive screening method that accurately estimates brain amyloid
burden would therefore fulfil a critical unmet need in the care of
the elderly.
The identification of blood-based biomarkers associated with
AD diagnosis [69] or distinct endophenotypes of AD pathology
such as brain atrophy [1012], hippocampal metabolite abnormalities [13] and amyloid burden [14], have previously been
reported. In these studies, proteomic analyses were combined with
neuroimaging methods to identify plasma signals associated with
measures of AD pathology. In this study, a different strategy was
used by examining the association between brain amyloid burden
and a panel of 146 plasma analytes proteins, complexes and
metabolites measured by Rules Based Medicine, Inc. (RBM)
(Austin, TX) using the Human Discovery Multi-Analyte Profile
(MAP) 1.0 panel and a Luminex 100 platform. Some of the
analytes on this panel, such as apolipoprotein E (APOE) and
Results
RBM analytes associate with Ab levels in the brain
Levels of analytes measured by the RBM Human Discovery
MAP 1.0 from ADNI plasma samples were compared to fibrillar
amyloid in the RBM-PiB PET cohort (N = 71). Characteristics of
this subcohort are summarised in Table 1 where it can be seen that
brain amyloid burden was almost significantly different at the 0.05
level between diagnostic groups (Kruskal-Wallis (KW) x2 test pvalue 0.055). The distribution of brain amyloid burden in the
RBM-PiB PET cohort is shown in Figure S1. In the slightly larger
ADNI-PiB PET cohort (i.e. all ADNI subjects with [11 C] PiB-PET
scans performed at baseline), whose sample characteristics are
shown in Table S1, brain amyloid burden was found to be
significantly different across diagnostic groups (KW p-value 0.022).
The analytes most associated with brain amyloid burden in the
RBM-PiB PET cohort, after taking into account co-variates (age,
gender, years of education, number of APOE E 4 alleles and the
number of days between [11 C]-PiB PET scan and plasma sample),
are shown in Table 2.
Prediction of brain amyloid burden using plasma RBM

analytes
To determine if a subset of the RBM panel was able to predict
fibrillar amyloid levels in the brain, multiple linear regression was
used. In this analysis the following subject co-variates were
included: age at plasma sample, years of education, gender, the
number of APOE E 4 alleles and the difference, in days, between
plasma sampling and [11 C]-PiB PET scan date. Multiple linear
Table 1. Characteristics of the ADNI RBM-PiB PET cohort.
Diagnostic group (number of subjects)

Characteristics
Control (3)
MCI (52)
AD (16)
P-value
Subject age in years at time of plasma sample
77.4 [5.6]
75.4 [11.1]
72.3 [8.2]
0.290
(Median [IQR])
Sex (Male/Female)
1/2
37/15
10/6
0.263
Years of education (Median [IQR])
13.0 [3.0]
16.0 [5.0]
16.0 [5.3]
0.398
Number of APOE E 4 alleles (0/1/2)
2/1/0
25/22/5
7/7/2
0.977
Days between [ C]-PiB PET scan and plasma sample (Median [IQR])
5.0 [15.5]
23.5 [60.5]
21.5 [42.3]
0.288
Average PiB uptake
1.31 [0.108]
1.98 [0.723]
1.90 [0.438]
0.055
11
(Median [IQR])
98
Characteristics of the ADNI RBM-PiB PET subcohort by diagnostic group. P-values were calculated for differences across diagnostic groups using a Kruskal-Wallis x2 test
for continuous characteristics and simulated contingency p-values for categorical characteristics.

Table 2. RBM analytes associated with brain amyloid burden.
RBM analyte
Gene name
Uniprot ID
Partial SRC with

Ab
P-value
Benjamini-Hochberg corrected pvalue
C-peptide
INS
P01308
20.310
0.010
0.351
Fibrinogen (a, b and c)
FG(A/B/G)
P02671 P02675
P02679
20.307
0.010
0.351
Alpha-1-antitrypsin
SERPINA1
P01009
20.302
0.012
0.351
Pancreatic polypeptide
PPY
P01298
20.296
0.014
0.351
Complement C3
C3
P01024
20.296
0.014
0.351
Vitronectin
VTN
P04004
20.295
0.014
0.351
von Willebrand factor
VWF
P04275
20.287
0.017
0.363
Cortisol
(NA)
(NA)
0.271
0.025
0.412
Serum amyloid p-component
APCS
P02743
20.268
0.027
0.412
AXL receptor tyrosine kinase
AXL
P30530
0.266
0.028
0.412
Interleukin-3
IL3
P08700
0.261
0.032
0.412
Interleukin-13
IL13
P35225
0.252
0.038
0.412
Matrix metalloproteinase-9 total
MMP9
P14780
20.250
0.040
0.412
APOE
APOE
P02649
20.248
0.042
0.412
Leptin
LEP
P41159
20.248
0.042
0.412
Immunoglobulin E (IgE)
(NA)
(NA)
20.243
0.046
0.424
Analytes with a partial SRC p-value of v0.05 are shown. Benjamini-Hochberg corrected p-values were calculated to take into account the comparisons against all RBM
analytes.
either PiB positive (brain amyloid burden w1.5) or PiB negative

(brain amyloid burden v1.5), based on a threshold used in
relevant literature [16]. Similarly, the subjects LOO CV predicted
brain amyloid burden, based on the 13 RBM and co-variate
model, was dichotomised into either predicted PiB positive
(predicted brain amyloid burden w1.5) or predicted PiB negative
(predicted brain amyloid burden v1.5). For each subject the
predicted and actual PiB classes (positive or negative) were
compared; the predicted PiB classes identified actual PiB positive
subjects with a sensitivity of 0.918 and a specificity of 0.545. It was
also found that removing all co-variates, except the number of
APOE E 4 alleles, from the 13 RBM and co-variate regression
model gave a similar but slightly improved predictive ability (LOO
CV R2 0.311).
regression was applied to predict brain amyloid burden using these

co-variates only, giving a leave one out (LOO) cross validation
(CV) R2 of 0.040. When brain amyloid burden was regressed to
just the number of APOE E 4 alleles this gave a LOO CV R2 of
0.123. Then multiple linear regression was applied to predict brain
amyloid burden from both RBM analytes and co-variates; the
analysis was restricted to the 16 RBM analytes that had a partial
Spearmans rank correlation (SRC) uncorrected p-value v0.05
(Table 2) resulting in a LOO CV R2 of 0.276.
Overfitting was reduced by grouping correlated variables and
selecting one RBM analyte to represent each group (Figure 1); first
all the 16 RBM analytes were used, then analytes were removed
from the model one by one, in the order determined by clustering,
and LOO CV repeated. The order of analyte removal was: (1) von
willebrand factor, (2) leptin, (3) serum amyloid p-component, (4)
vitronectin, (5) interleukin-13, (6) component C3, (7) matrix
metalloproteinase-9 total, (8) immunoglobulin E (IgE), (9) APOE,
(10) pancreatic polypeptide, (11) alpha-1-antitrypsin, (12) interleukin-3, (13) cortisol, (14) fibrinogen and finally (15) AXL receptor
tyrosine kinase.
The grouping that resulted in the highest LOO CV R2 over all
possible hierarchical clustering cut-offs was used (Figure 2). This
was achieved when 13 RBM analytes were used: c-peptide,
fibrinogen, alpha-1-antitrypsin, pancreatic polypeptide, complement C3, vitronectin, cortisol, AXL receptor tyrosine kinase,
interleukin-3, interleukin-13, matrix metalloproteinase-9 total,
APOE and IgE (LOO CV R2 0.310, permutation test p-value
4610{5 ). The 13 RBM analyte and co-variate model is able to
account for approximately a third of the variance of brain amyloid
burden. The relative importance of variables to the model is
shown in Figure 3; the number of APOE E 4 alleles was seen to be
the most important variable for the model, but RBM analytes
contribute more to the model than years of education or age. The
71 subjects brain amyloid burden was then dichotomised into
Markers of brain amyloid burden associate with other

measures of AD pathology
The 16 RBM analytes found to associate with brain amyloid
burden in the ADNI RBM-PiB PET cohort were then tested for
association with other phenotypes known to relate to AD
pathology in the ADNI-RBM cohort (sample characteristics
shown in Table S2). In this analysis age, gender, years of
education and the number of APOE E 4 alleles were included in
partial correlation analysis. Subjects with missing values for a
relevant comparison were excluded. First the level of the markers
of brain amyloid burden were compared with the level of Ab1{42
in the CSF. Leptin was found to associate with CSF Ab1{42
(partial SRC 0.183, BH MTC p-value 0.0183). Additionally, when
multiple testing was not taken into account, vitronectin was
associated with CSF Ab1{42 (partial SRC 0.131, uncorrected pvalue 2.07610{2 ).
Markers of brain amyloid burden were then compared to AD
relevant brain regions as measured by structural magentic
resonance imaging (sMRI). The volume of the left and right
3
99
Figure 1. Analyte correlations. Heatmaps of partial SRC between RBM analytes significantly associated (p-value v0.05) with brain amyloid
burden, taking into account: age, gender, years of education, number of APOE E 4 alleles and the difference, in days, between plasma sampling and
[11 C]-PiB PET scan date. RBM analytes have been ordered by hierarchical clustering, the final cut-off is shown in purple. Variables that have been
grouped are shown in blue boxes, and the representative for each group is shown in a red box.
hippocampi, and the thickness of the left and right entorhinal

cortices were used, as these regions are known to be important in
AD pathology [12]. Alpha-1-antitrypsin was found to associate at
the 5% significance level with the thickness of both the left (partial
SRC 20.132, BH MTC p-value 0.0289) and right (partial SRC
20.145, BH MTC p-value 9.82610{3 ) entorhinal cortices.
Additionally, leptin was found to associate with the thickness of
the right entorhinal cortex (partial SRC 0.124, BH MTC p-value
0.0264). Cortisol was found to associate at the 5% significance
level with the volume of both the left (partial SRC 20.158, BH
MTC p-value 2.71610{3 ) and right hippocampi (partial SRC
20.161, BH MTC p-value 2.07610{3 ).
Different cognitive tests, such as the Mini Mental State Exam
(MMSE) and Alzheimers Disease Assessment Scale-cognitive
subscale (ADAS-cog) 13, assess different aspects of cognitive
decline. The association between scores from these tests and levels
of markers of brain amyloid burden was analysed; 4/16 of the
markers of brain amyloid burden alpha-1-antitrypsin (partial
SRC 20.145, BH MTC p-value 3.02610{3 ), complement C3
(partial SRC 20.189, BH MTC p-value 9.46610{5 ), cortisol
(partial SRC 20.162, BH MTC p-value 8.33610{4 ) and
fibrinogen (partial SRC 20.130, BH MTC p-value 8.28610{3 )
were found to associate with the total MMSE score at the 5%
level. In the same cohort (ADNI-RBM), three of these markers

alpha-1-antitrypsin (partial SRC 0.172, BH MTC p-value 0.0403),
complement C3 (partial SRC 0.119, BH MTC p-value 0.0467)
and fibrinogen (partial SRC 0.111, BH MTC p-value 6.34610{4 )
were found to associate with the ADAS-cog 13 score at the 5%
significance level.
Levels of the 16 markers of brain amyloid burden were then
compared between different diagnostic groups (control, MCI and
AD) in the various cohorts, to assess whether these markers were
related to clinical diagnosis. It was found that half of the
biomarkers of brain amyloid burden measured APOE (BH
MTC KW p-value 1.24610{9 , AD/control median difference
(MD) 20.143), complement C3 (BH MTC KW p-value
8.88610{8 , MCI/control MD 20.0483), cortisol (BH MTC
KW p-value 2.40610{3 , AD/control MD 0.0366), interleukin-3
(BH MTC KW p-value 8.48610{3 , AD/control MD 20.0670),
leptin (BH MTC KW p-value 8.48610{3 , AD/control MD
20.112), pancreatic polypeptide (BH MTC KW p-value
1.92610{2 , AD/control MD 0.122), alpha-1-antitrypsin (BH
MTC KW p-value 1.75610{7 , AD/control MD 0.0300) and
vitronectin (BH MTC KW p-value 5.50610{3 , AD/control MD 0.0344) significantly differ at the 5% level between diagnostic
groups.
100

shown that the difference in plasma APOE levels between AD and

control subjects is largely driven by the APOE genotype, and so the
interaction of these two variables was studied further. In the
ADNI-RBM cohort the number of APOE E 4 alleles was seen to
have a negative effect on plasma APOE levels (KW p-value
v2.20610{16 , Table 3). The negative effect of APOE genotype on
plasma APOE levels in this ADNI subcohort was demonstrated
recently using Analysis of Variance (ANOVA) [20]. The analysis
presented here shows that this result holds when assumptions of
normality are dropped. The negative effect of the number of APOE
E 4 alleles on plasma APOE levels observed fits with the findings of
some literature [15,17,18], but is the opposite of the positive effect
seen by both Evans et al. and Thambisetty et al. (2010) [14,19].
Given the discrepancy between this finding and those in two
published studies, the relationship between the number of APOE E
4 alleles and plasma APOE levels was studied in an independent
cohort of 694 subjects (AddNeuroMed and Kings Health Partners
Dementia Case Register, ANM + KNPDCR). In the ANM +
KHPDCR cohort the number of APOE E 4 alleles was also seen to
have a negative effect on plasma APOE levels (KW p-value
v2.20610{16 , Table 3). Both studies that have found a positive
relationship between APOE E 4 alleles on plasma APOE levels have
been conducted in cohorts of cognitively normal subjects [14,19],
which may account for this inconsistency. However, a similar
negative effect was found in the control subjects (who are
cognitively normal) of the ANM + KHPDCR cohort (KW pvalue 1.57610{4 , Table 3). This suggests another factor, other
than cognitive decline, is responsible for the discrepancies between
these studies.
Given that the number of APOE E 4 alleles affects both plasma
APOE levels and brain amyloid burden, it is possible that the
number of APOE E 4 alleles confounds the association between
plasma APOE levels and brain amyloid burden. To test this,
partial correlation analysis was repeated excluding the number of
APOE E 4 alleles, this increased the correlation of plasma APOE
level and brain amyloid burden (partial SRC 20.393, p-value
6.9610{4 ), which indicated that the association is indeed partly
confounded by the number of APOE E 4 alleles.
Figure 2. Cross validation of multiple linear regression. Barplot

of the LOO CV R2 of multiple linear regressions of brain amyloid burden
against a range of RBM analytes and co-variates. Subsets of RBM
analytes that associated with brain amyloid levels with a p-value v0.05
were used. Various subsets were chosen by hierarchical clustering at
various cutoffs, with the analyte most associated with brain amyloid
burden in each cluster chosen to represent that cluster. Age, gender,
years of education, the number of APOE E 4 alleles and the difference, in
days, between plasma sampling and [11 C]-PiB PET scan were used as covariates.
The effect of APOE genotype on APOE level in plasma

and brain amyloid burden
The number of APOE E 4 alleles and the level of APOE in
plasma were the two variables that contributed most to the
regression model. A number of studies have shown that APOE
level in plasma is affected by presence of APOE E 4 alleles
[14,15,1719]. Additionally, Slooter et al., [17] have previously
Discussion
In this study, fibrillar amyloid beta levels, in ADNI subjects,
have been compared and related to the level of analytes on the
RBM panel in plasma. Brain amyloid burden appears to be
distributed bimodally in the RBM-PiB PET cohort, as has been
previously reported for a larger ADNI subcohort by Ewers et al.
[21]. Associations of C-peptide, fibrinogen, alpha-1-antitrypsin,
pancreatic polypeptide, complement C3, vitronectin, von willebrand factor, cortisol, serum amyloid p-component, AXL
receptor tyrosine kinase, interleukin-3, interleukin-13, matrix
metallproteinase-9, APOE, leptin and immunoglobulin E with
brain amyloid burden have been found in this study. Some of these
markers of brain amyloid burden were also found to associate with
other AD related phenotypes, such as CSF Ab1{42 , MRI features,
cognitive tests and diagnostic groups. In regression models it was
found that models including both RBM analytes and co-variates
performed better than those using only co-variate information,
suggesting that the RBM panel of analytes can be used as markers
of brain amyloid burden. Combining highly correlated variables
was found to reduce overfitting and led to a set of 13 RBM
analytes that together with co-variates could explain w30% of the
variance of brain amyloid burden and predict PiB positive
individuals with a high sensitivity. This result, and the increased
predictive accuracy of the 13 RBM model in comparison to using 101
Figure 3. Relative importance in multiple linear regression.

Relative importance of variables used in multiple linear regression
assessed using the LMG relative importance score [51]. When all the
data is used to fit the model it explains 62.0% of the variance of the
brain amyloid burden, here the contribution of the variables used in this
model are shown.
Table 3. The effect of APOE genotype on plasma APOE levels.
The plasma level in log mg/ml of APOE in

subjects with n APOE E 4 alleles (median [IQR])
Cohort
n=0
ADNI-RBM
1.79 [0.200]
ANM + KHPDCR
1.91 [9.75610
n=1
1.66 [0.203]
22
Kruskal-Wallis x2 Pvalue
n=2
,2.20610216
1.52 [0.233]
1.88 [9.00610
22
1.82 [7.00610
22
,2.20610216
ANM + KHPDCR
controls
1.9 [0.100]
1.88 [8.0061022]
1.83 [3.0061022]
1.5761024
Level of plasma APOE stratified by the subjects number of APOE E 4 alleles.

102
than amyloid pathology, and so it is possible that the lack of

association of some of the markers with, for example, diagnostic
groups is due to the mixture of high and low brain amyloid burden
control subjects used. However, the lack of association of the
majority of the markers with CSF Ab1{42 levels is of greater
concern because CSF Ab1{42 levels are strongly associated with
brain amyloid burden [4,5]; this may indicate that we are overfitting the available data and further highlights the need for
datasets with larger sample sizes for future studies of markers of
brain amyloid burden.
While half of the markers associated with diagnostic groups, 3/8
of these markers alpha-1-antitrypsin, pancreatic polypeptide and
interleukin-3 had a median difference between AD (or MCI) and
control subjects that was of the opposite sign to the partial SRC
coefficient measuring their association with brain amyloid burden.
This result is surprising as brain amyloid burden is positively
associated with AD and MCI diagnosis groups. This discrepancy
could relate to the delay between these disease stages, and may
mean that the level of some of the markers in plasma changes
during Ab deposition and then changes again, but in the opposite
direction, before the onset of clinical symptoms. Similar u-shaped
profiles, but between subjects in different diagnostic groups, have
been observed (cross-sectionally) in the level of many leukocyte
transcripts during AD progression [22].
Partial correlation showed that the number of APOE E 4 alleles
partly confounded the association between APOE level in plasma
and brain amyloid burden; however, plasma APOE levels did help
a regression model predict brain amyloid burden and so further
study is required to get a clearer idea of the APOE E 4 independent
information conveyed by plasma APOE levels. This study has
revealed many novel potential markers of brain amyloid burden,
chosen to give APOE E 4 independent information, as well as
replicating findings from other studies. This will allow further
validation work that can test the replicability and clinical utility of
these markers.
In a previous study that used discovery proteomics to identify
proteins associated with brain amyloid levels, Thambisetty et al.
(2010) showed that levels of APOE and Complement C3 precursor
in plasma were different between subjects with high and low brain
amyloid burdens [14]. It was encouraging that both were seen to
be associated with brain amyloid burden in this study as well.
Complement C3 precursor has also been found to be associated
with atrophy of hippocampal volume, another imaging marker of
AD [11], and to have a role in plaque clearance in a mouse model
[23]. It has also been found along with vitronectin to be at
different levels in serum between control and AD subjects [6]. The
level of fibrinogen gamma was also found to be associated with
only co-variates to predict fibrillar amyloid alone, indicate that

these analytes reflect levels of fibrillar amyloid in the brain.
The potential of APOE level in plasma to be used as a
biomarker for brain amyloid burden, previously shown in
Thambisetty et al. (2010) [14], is given support by this study.
However it should be noted that the association between plasma
APOE level and brain amyloid burden was seen to be positive in
that study and negative in this. This inconsistency may relate to
differences between the cohorts, for example the BLSA cohort
studied in Thambisetty et al. (2010) were selected to be cognitively
normal. This fits with the recent finding that plasma level of
APOE correlates negatively with brain amyloid burden in the
Australian Imaging, Biomarker and Lifestyle Flagship Study of
Ageing, which includes subjects suffering from AD [15]. Similarly,
the effect of the presence of APOE E 4 alleles on plasma APOE
level in the study presented here was found to be the opposite of
that found in both Thambisetty et al. (2010) and Evans et al.
[14,19], however it is the same as that seen in Slooter et al., Siest
et al. and Gupta et al. [15,17,18]. Additionally, this relation was
replicated in an independent cohort (ANM + KHPDCR) and its
control subcohort. This latter finding suggests that the differences
between the findings of these studies is not due to the subjects
cognitive status. The factor/s responsible for these inconsistencies
are not known, but the fact that strong associations are seen in all
studies is encouraging.
While no association between the level of RBM analytes in
plasma and brain amyloid burden was found to be significant at
the 5% level after multiple testing corrections, it should be noted
the cohort used (RBM-PiB PET) contained relatively few subjects
and that many analytes that associated significantly at the
uncorrected 0.05 level have known relations to AD pathology,
for example levels of APOE and complement C3 in plasma have
previously been found to associate with fibrillar amyloid levels in
the Baltimore Longitudinal Study of Aging (BLSA) cohort [14]. In
addition, some of these analytes were subsequently found to
associate with surrogate phenotypes of AD pathology in the larger
ADNI-RBM cohort, such as: diagnostic groups, MMSE score,
ADAS-cog 13 score, CSF Ab1{42 level, and the thickness and/or
volume of the entorhinal cortices. However, 7/16 of the markers
of brain amyloid burden c-peptide, von Willebrand factor, serum
amyloid p-component, AXL receptor tyrosine kinase, interleukin13, matrix metalloproteinase-9 total and IgE were not found to
associate with any of the surrogate phenotypes of AD pathology
that were tested. It should be noted that 5 of the 7 are linked to AD
in the literature (Table S3), only IgE and interleukin-13 have no
prior reported association. Most of these surrogate phenotypes,
except CSF Ab1{42 , are believed to change at a later disease stage

atrophy of hippocampal volume in Thambisetty et al. (2011) [11].

Fibrinogen alpha, beta and gamma are targeted by the same RBM
analyte and were found to associate with brain amyloid burden in
this study.
Of the 16 RBM analytes whose level in plasma associated with
brain amyloid burden, many have known relationships with
Alzheimers disease. The levels of the following have previously
been found to be different between control and AD subjects:
alpha-1-antitrypsin [9,24], APOE [9], cortisol [9,25,26], interleukin-3 [27], matrix metalloproteinase-9 [9,28], pancreatic polypeptide [9,29], serum amyloid p-component [30] and von Willebrand
factor [31]. Serum amyloid p-component [32] and insulin [33]
have been shown to affect AD-like pathology in vitro. Interleukin3 [34] and leptin [35] have been found to affect the interaction of
neurons and Ab. Additionally, interleukin-13 has been found to be
produced in microglia in response to Ab [36]. More recently,
APOE and matrix metalloproteinase-9 have been shown to be
involved together in the breakdown of the blood brain barrier,
which can initiate neurodegeneration [37].
Given the relatively small number of subjects in this study it was
not practical to separate the subjects into training and test sets, to
assess the predictive accuracy of the regression model. Instead, kfold cross-validation was used, allowing more of the subjects to be
used for fitting the regression model. Generally it is advisable to
use 10-fold cross-validation because it has been found to have a
lower variance [38]. However, given the limited number of
samples available, a leave one out cross-validation approach was
used in this study to allow the maximal use of the subjects
available. Given the limited number of subjects on which the
model is based, it will be important in the future to study the
ability of these biomarkers to predict brain amyloid burden in an
independent cohort. Validation studies would benefit from greater
numbers of subjects and better sampling strategies. For example,
the distribution of brain amyloid burden in the RBM-PiB PET
subcohort is affected by the sampling strategy applied to select
control subjects for RBM measurements; only plasma of control
subjects with high CSF Ab1{42 were selected, for reasons
unrelated to the current study. This means that the resulting
model may not extrapolate to cognitively normal subjects with
high brain amyloid burden, this could be tested in a validation
study. Additionally, the use of only three control subjects may
make the regression model less likely to generalise to prediction of
brain amyloid burden in early Alzheimers disease.
In conclusion, analytes associating with brain amyloid burden
have the potential to act as biomarkers of early AD-related
pathology. In this study sixteen analytes were found to associate
with brain amyloid burden, including two (APOE and complement C3) that had had already been shown to associate with brain
amyloid burden in an independent cohort. Some of these analytes
were also found to associate with other AD related phenotypes in a
larger ADNI subcohort, such as: CSF Ab1{42 , MRI features,
cognitive scores and diagnostic groups. Some of these analytes
were found to correlate highly with each other, and so a
representative set of thirteen analytes c-peptide, fibrinogen,
alpha-1-antitrypsin, pancreatic polypeptide, complement C3,
vitronectin, cortisol, AXL receptor kinase, interleukin-3, interleukin-13, matrix metalloproteinase-9 total, APOE and IgE were
used along with subject age, gender, years of education, the
number of APOE E 4 alleles and sampling dates to predict brain
amyloid burden. The 13 analyte and co-variate model was found
by cross-validation to account for w30% of the variance of brain
amyloid burden, as opposed to * 413% using just co-variates
alone, showing the potential of plasma analytes as markers of brain
amyloid burden. The model was also able to predict PiB positive
individuals with a high sensitivity. The two variables with the

largest contribution to the model were found to be the number of
APOE E 4 alleles and plasma APOE level. The association of
plasma APOE level with brain amyloid burden was shown to be
partly confounded by the number of APOE E 4 alleles, highlighting
the importance of novel biomarkers that are less confounded by
the APOE genotype revealed by this study.

Ethics statement
Written informed consent was obtained from all participants in
ADNI and the study was conducted with prior institutional ethics
approval. Both ANM and KHPDCR were approved by the South
London and Maudsley NHS Foundation Trust ethics committee.
Ethics committee approval was also obtained at each of the
participating centres in accordance with the Alzheimers Associations published recommendations [39].
ADNI Data
Data used in the preparation of this article were obtained from
the ADNI database (adni.loni.ucla.edu). The ADNI was launched
in 2003 by the National Institute on Aging (NIA), the National
Institute of Biomedical Imaging and Bioengineering (NIBIB), the
Food and Drug Administration (FDA), private pharmaceutical
companies and non-profit organisations, as a $60 million, 5- year
public-private partnership. The primary goal of ADNI has been to
test whether serial magnetic resonance imaging (MRI), positron
emission tomography (PET), other biological markers, and clinical
and neuropsychological assessment can be combined to measure
the progression of mild cognitive impairment (MCI) and early
Alzheimers disease (AD). Determination of sensitive and specific
markers of very early AD progression is intended to aid
researchers and clinicians to develop new treatments and monitor
their effectiveness, as well as lessen the time and cost of clinical
trials.
The Principal Investigator of this initiative is Michael W.
Weiner, MD, VA Medical Center and University of California
San Francisco. ADNI is the result of efforts of many coinvestigators from a broad range of academic institutions and
private corporations, and subjects have been recruited from over
50 sites across the U.S. and Canada. The initial goal of ADNI was
to recruit 800 adults, ages 55 to 90, to participate in the research,
approximately 200 cognitively normal older individuals to be
followed for 3 years, 400 people with MCI to be followed for 3
years and 200 people with early AD to be followed for 2 years. For
up-to-date information, see www.adni-info.org.
Demographic (age, gender, years of education), genetic (number
of APOE E 4 alleles), diagnosis (control, MCI or AD at a given date)
and analyte (metabolite/protein/complex) levels in plasma were
compared with brain amyloid burden (or other markers of AD
pathology such as: CSF Ab1{42 level, MRI features, cognitive
scores or diagnostic groups). Diagnoses were recorded for each
subject at each visit. Plasma and CSF were collected from fasted
subjects using the procedures described previously [29,40]. Levels
of 190 analytes were measured from subject plasma using the
Rules Based Medicine (RBM, rulesbasedmedicine.com, Austin,
TX) Human Discovery Multi-Analyte Profile (MAP) 1.0 panel and
a Luminex 100 platform [41]. Measurements of 44 analytes were
excluded on the basis of quality control, leaving 146 analytes in the
subsequent analysis. The levels of all 146 analytes (except:
apolipoprotein H, complement factor H, E selectin, epidermal
growth factor, fibrinogen, interleukin-12 subunit p40, placenta
growth factor, serum glutamic oxaloacetic transaminase and 103
7
PET scan and plasma sample. In the case of correlations between

RBM analytes and CSF Ab, ADAS-cog 13 or MRI features,
subjects whose relevant data was missing were excluded. RBM
analytes were clustered based upon (1 their partial SRC) using
the R function hclust in the stats package with default settings, and
displayed using function heatmap.2 from the R gplots package
[49].
thrombopoietin) were log transformed to improve the fit of the

levels to the normal distribution. Description of methods used to
derive measurements of regional [11 C] PiB-PET levels have been
given in Jagust et al. (2010) and (2011) previously [16,42]. In this
study a similar approach was taken by averaging over regional
[11 C] PiB-PET measurements in parietal, frontal, anterior
cingulate and precuneus regions of interest, to derive a global
measure of brain amyloid burden.
Data from either baseline or 12 months were used for these
cohorts, as described above, chosen to increase the number of
subjects with available data. Eighty four subjects (ADNI-PiB PET
cohort) had a [11 C] PiB-PET scan 12 months after baseline, the
characteristics of this sample are shown in Table S1. Five hundred
and sixty six subjects (ADNI-RBM cohort) had RBM analytes
measured in plasma collected at baseline, the characteristics of this
sample at baseline is shown in Table S2. Seventy one subjects
(RBM-PiB PET cohort) had RBM analytes measured in plasma
collected 12 months after baseline and a [11 C] PiB-PET scan
within a year of this, the characteristics of this sample at the date of
plasma collection is shown in Table 1. Converters from control to
MCI between plasma sample and [11 C] PiB-PET scan date
account for the discrepancy in diagnostic groups between the
ADNI-PiB PET and RBM-PiB PET cohorts. Subject age was
determined based on the same dates.
Sample characteristics of each cohort by diagnostic group were
analysed in R [43]. Many continuous variables were not
distributed normally, and so are described by median and
interquartile range instead of mean and standard deviation, this
was also the reason that non-parametric statistical tests were used.
Continuous characteristics were tested for differences over
diagnostic groups by the non-parametric Kruskal-Wallis x2 test,
using kruskal.test in the R stats package. Discrete characteristics
were tested for differences over diagnostic groups by simulated
contingency table p-values, in the fisher.test function in the R stats
package, using 2,000 Monte Carlo samples.
Linear regression
Linear regression was performed using the lm function in the R
stats package. This was appropriate because although many
variables were not distributed normally, the residuals of the
regression models used were approximately. Before regression,
measurements of each RBM analyte were transformed to a
standard deviation of one to allow each analyte to have equal
influence on the model (but not transformed to a mean of zero, to
make the analysis more comparable with that used in Thambisetty
et al. (2010) [14]). LOO CV was performed by fitting linear
regression models to the data, leaving out one subject at a time,
and using the model to predict brain amyloid burden in that
subject based on the fitted model. LOO CV R2 was calculated as
the square of the Pearsons correlation coefficient, calculated using
cor.test, between the predicted and observed brain amyloid
burden.
A cut-off of 1.5 was used to dichotomise brain amyloid burden
in the RBM-PiB PET cohort as PiB positive (49 subjects) or
negative (22 subjects), as previously suggested by Jagust et al.
(2010) [16]. Predicted values were similarly dichotomised.
Sensitivity and specificity of this prediction was calculated in R
using epi.test in the epiR package [50].
Permutation tests were performed by permuting brain amyloid
burden across subjects in the data, and re-calculating the LOO
CV R2 that was achieved by fitting the regression model to the
resulting data. 100,000 permutations were used, and the number
of cases in which the LOO CV R2 exceeded that achieved with
the original dataset was recorded. Relative importance was
calculated using the LMG score [51], using the R package
relaimpo [52].
MRI scan analysis

Dicom format MRI data was downloaded from the ADNI
website (www.loni.ucla.edu/ADNI). Data from 1.5 T scanners
was used with data collected from a variety of MR-systems with
protocols optimised for each type of scanner. The MRI protocol
included a high resolution sagittal 3D T1-weighted MPRAGE
volume (voxel size 1.161.161.2 mm3 ) acquired using a custom
pulse sequence specifically designed for the ADNI study to ensure
compatibility across scanners. Full brain and skull coverage was
required for the MRI datasets and detailed quality control carried
out on all MR images according to previously published quality
control criteria [44,45].
We applied the Freesurfer pipeline (version 4.5.0) to the MRI
images to produce regional cortical thickness and volumetric
measures as previously described [46] to produce hippocampal
and entorhinal cortex volumes, as well as entorhinal cortical
thickness. All volumetric measures from each subject were
normalised by the subjects intracranial volume. Cortical thickness
measures were not normalised [47] and were used in their raw
form.
Independent cohort data for validation of the effect of

APOE E 4 genotype on APOE level in plasma
EDTA plasma samples from fasted subjects were obtained from
two independent cohorts: ANM a multicentre European study
across six centres [53] and KHPDCR a UK based study. The
combined cohort contained 269 control, 163 MCI and 262 AD
subjects. APOE genotype was determined using DNA extracted
from blood leukocytes by a standard phenol-chloroform extraction. The three main alleles APOE E 2, APOE E 3 and APOE E 4
differ at two residues, so consist of a two single nucleotide
polymorphism (SNP) haplotype. The SNPs rs429358 and rs7412
were genotyped and the allele inferred. SNPs were determined by
allelic discrimination assays based on fluorogenic 59 nuclease
activity. TaqMan SNP genotyping assays were performed on an
ABI Prism 7900HT and analyzed using SDS software, according
to the manufacturers instructions (Applied Biosystems, Warrington, UK). 199 control, 103 MCI and 120 AD subjects were found
to have 0 APOE E 4 alleles. 65 control, 56 MCI and 112 AD
subjects were found to have 1 APOE E 4 alleles. 5 control, 4 MCI
and 30 AD subjects were found to have 2 APOE E 4 alleles. The
Human Neurodegenerative Panel 1 (7-plex) Cat. HNDG1-36K
MILLIPLEX MAP multiplex panels, developed by Merck
Millipore, was used to measure APOE level in plasma.
Correlation analysis
104
R was used to analyse SRC and partial SRC, using cor.test in

the stats package and pcor.test in the ppcor package [48]
respectively. Partial correlations were used to take into account
subjects: age, gender, years of education, number of APOE E 4
alleles and the number of days separating the date of [11 C]-PiB

Table S3 Association of markers with AD phenotypes.
(PDF)
Figure S1 Distribution of brain amyloid burden. A
stacked histogram showing the distribution of brain amyloid

burden for different diagnostic groups. Control (dark blue), MCI
(green) and AD (red) represent subjects who remained in these
diagnostic groups throughout follow up period. Control/MCI
(light blue) and MCI/AD (orange) represents subjects whose
diagnosis converted between these groups during the follow up
period. Brain amyloid burden is in relative units.
(TIF)
Acknowledgments
We are grateful to peer-reviewers, whose suggestions helped us to refine
this manuscript.
Data used in preparation of this article were obtained from the
Alzheimers Disease Neuroimaging Initiative (ADNI) database (adni.loni.ucla.edu). As such, the investigators within the ADNI contributed to the
design and implementation of ADNI and/or provided data but did not
participate in analysis or writing of this report. A complete listing of ADNI
investigators can be found at: http://adni.loni.ucla.edu/wp-content/
uploads/how_to_apply/ADNI_Acknowledgement_List.pdf.
Table S1 Characteristics of the ADNI-PiB PET cohort
by diagnostic group. P-values were calculated when appropriate for differences across diagnostic groups, using a Kruskal-Wallis
x2 test for continuous characteristics and simulated contingency
table p-values for discrete characteristics.
(PDF)
Conceived and designed the experiments: M. Thambisetty AS SL SN RD.
Analyzed the data: SJK. Wrote the paper: SJK M. Thambisetty AS SN
RD. APOE measurements in AddNeuroMed and DCR: JRC AH IP MW
MKL KL. Input on image analysis: EW. Provided cluster computing
support: CJ. AddNeuroMed clinical centre lead: HS IK M. Tsolaki BV
PM.
Characteristics of the ADNI-RBM cohort by

diagnostic group. P-values were calculated when appropriate
for differences across diagnostic groups, using a Kruskal-Wallis x2
test for continuous characteristics and simulated contingency table
p-values for discrete characteristics.
(PDF)
Table S2
References
1. Golde TE, Schneider LS, Koo EH (2008) Anti-Ab therapeutics in Alzheimers
disease: the need for a paradigm shift. Neuron 69: 203213.
2. Reiman EM, Langbaum JB, Fleisher AS, Caselli RJ, Chen K, et al. (2011)
Alzheimers Prevention Initiative: a plan to accelerate the evaluation of
presymptomatic treatments. J Alzheimers Dis 26: 321329.
3. Aisen PS, Andrieu S, Sampaio C, Carrillo M, Khachaturian ZS, et al. (2011)
Report of the task force on designing clinical trials in early (predementia) AD.
4. Grimmer T, Riemenschneider M, Forstl H, Henriksen G, Klunk WE, et al.
(2009) Beta amyloid in Alzheimers disease: increased deposition in brain is
reected in reduced concentration in cerebrospinal fluid. Biol Psychiatry 65: 927
934.
5. Koivunen J, Pirttila T, Kemppainen N, Aalto S, Herukka SK, et al. (2008) PET
amyloid ligand [11C] PIB uptake and cerebrospinal uid beta-amyloid in mild
cognitive impairment. Dement Geriatr Cogn Disord 26: 378383.
6. Zhang R, Barker L, Pinchev D, Marshall J, Rasamoelisolo M, et al. (2004)
Mining biomarkers in human sera using proteomic tools. Proteomics 4: 244
256.
7. Hye A, Lynham S, Thambisetty M, Causevic M, Campbell J, et al. (2006)
Proteome-based plasma biomarkers for Alzheimers disease. Brain 129: 3042
3050.
8. Soares HD, Potter WZ, Pickering E, Kuhn M, Immermann FW, et al. (Ahead of
print) Plasma biomarkers associated with the apolipoprotein e genotype and
alzheimer disease. Arch Neurol.
9. Doecke JD, Laws SM, Faux NG, Wilson W, Burnham SC, et al. (Ahead of print)
Blood-based protein biomarkers for diagnosis of Alzheimer disease. Arch
Neurol.
10. Thambisetty M, Simmons A, Velayudhan L, Hye A, Campbell J, et al. (2010)
Association of plasma clusterin concentration with severity, pathology, and
progression in Alzheimer disease. Arch Gen Psychiatry 67: 739748.
11. Thambisetty M, Simmons A, Hye A, Campbell J, Westman E, et al. (2011)
Plasma biomarkers of brain atrophy in Alzheimers disease. PloS one 6: e28527.
12. Thambisetty M, An Y, Kinsey A, Koka D, Saleem M, et al. (2012) Plasma
clusterin concentration is associated with longitudinal brain atrophy in mild
cognitive impairment. Neuroimage 59: 212217.
13. Thambisetty M, Hye A, Foy C, Daly E, Glover A, et al. (2008) Proteome-based
identification of plasma proteins associated with hippocampal metabolism in
early Alzheimers disease. J Neurol 255: 17121720.
14. Thambisetty M, Tripaldi R, Riddoch-Contreras J, Hye A, An Y, et al. (2010)
Proteome-based plasma markers of brain amyloid-b deposition in non-demented
older individuals. J Alzheimers Dis 22: 10991109.
15. Gupta VB, Laws SM, Villemagne VL, Ames D, Bush AI, et al. (2011) Plasma
apolipoprotein E and Alzheimer disease risk: the AIBL study of aging.
Neurology 76: 10911098.
16. Jagust WJ, Bandy D, Chen K, Foster NL, Landau SM, et al. (2010) The
Alzheimers Disease Neuroimaging Initiative positron emission tomography
core. Alzheimers Dement 6: 221229.
17. Slooter AJ, de Knijff P, Hofman A, Cruts M, Breteler MM, et al. (1998) Serum
apolipoprotein E level is not increased in Alzheimers disease: the Rotterdam
study. Neurosci Lett 248: 2124.
18. Siest G, Bertrand P, Qin B, Herbeth B, Serot JM, et al. (2000) Apolipoprotein E
polymorphism and serum concentration in Alzheimers disease in nine European
centres: the ApoEurope study. Clin Chem Lab Med 38: 721730.
19. Evans AE, Zhang W, Moreel JF, Bard JM, Ricard S, et al. (1993)
Polymorphisms of the apolipoprotein B and E genes and their relationship to
plasma lipid variables in healthy Chinese men. Hum Genet 92: 191197.
20. Johnstone D, Milward EA, Berretta R, Moscato P, ADNI (2012) Multivariate
protein signatures of pre-clinical Alzheimers disease in the Alzheimers Disease
Neuroimaging Initiative (ADNI) plasma proteome dataset. PLoS one 7: e34341.
21. Ewers M, Insel P, Jagust WJ, Shaw L, Trojanowski JQ, et al. (2012) CSF
Biomarker and PIBPETDerived Beta-Amyloid Signature Predicts Metabolic,
Gray Matter, and Cognitive Changes in Nondemented Subjects. Cerebral
Cortex 22: 19932004.
22. Lunnon K, Ibrahim Z, Proitsi P, Lourdusamy A, Newhouse S, et al. (2012)
Mitochondrial Dysfunction and Immune Activation are Detectable in Early
Alzheimers Disease Blood. J Alz Dis 30: 685710.
23. Maier M, Peng Y, Jiang L, Seabrook TJ, Carroll MC, et al. (2008) Complement
C3 deficiency leads to accelerated amyloid b plaque deposition and
neurodegeneration and modulation of the microglia/macrophage phenotype
in amyloid precursor protein transgenic mice. J Neurosci 28: 63336341.
24. Wetterling T, Tegtmeyer KF (1994) Serum alpha 1-antitrypsin and alpha 2macroglobulin in Alzheimers and Binswangers disease. Clin Investig 72: 196
199.
25. Bemelmans KJ, Noort A, Rijk RD, Middelkoop HAM, Kempen GMJV, et al.
(2007) Plasma cortisol and norepinephrine in Alzheimers disease: opposite
relations with recall performance and stage of progression. Acta Neuropsychiatr
19: 231237.
26. Lee BK, Glass TA, Wand GS, McAtee MJ, Bandeen-Roche K, et al. (2008)
Apolipoprotein E genotype, cortisol, and cognitive function in communitydwelling older adults. Am J Psychiat 165: 14561464.
27. Ray S, Britschgi M, Herbert C, Takeda-Uchimura Y, Boxer A, et al. (2007)
Classification and prediction of clinical Alzheimers diagnosis based on plasma
signaling proteins. Nat Med 13: 13591362.
28. Lorenzl S, Albers DS, Relkin N, Ngyuen T, Hilgenberg SL, et al. (2003)
Increased plasma levels of matrix metalloproteinase-9 in patients with
Alzheimers disease. Neurochem Int 43: 191196.
29. OBryant SE, Xiao G, Barber R, Huebinger R, Wilhelmsen K, et al. (2011) A
Blood-Based Screening Tool for Alzheimers Disease That Spans Serum and
Plasma: Findings from TARC and ADNI. PLoS one 6: e28092.
30. Nishiyama E, Iwamoto N, Kimura M, Arai H (1996) Serum amyloid P
component level in Alzheimers disease. Dementia 7: 256259.
31. OBryant SE, Xiao G, Barber R, Reisch J, Doody R, et al. (2010) A serum
protein-based algorithm for the detection of Alzheimer disease. Arch Neurol 67:
10771081.
105
44. Simmons A, Westman E, Muehlboeck S, Mecocci P, Vellas B, et al. (2009) MRI

measures of Alzheimers disease and the AddNeuroMed study. Ann NY Acad Sci
1180: 4755.
45. Simmons A, Westman E, Muehlboeck S, Mecocci P, Vellas B, et al. (2011) The
AddNeuroMed framework for multi-centre MRI assessment of longitudinal
changes in Alzheimers disease : experience from the first 24 months. Int J Ger
Psych 26: 7582.
46. Westman E, Muehlboeck JS, Simmons A (2012) Combining MRI and CSF
measures for classification of Alzheimers disease and prediction of mild
cognitive impairment conversion. Neuroimage 62: 229238.
47. Westman E, Aguilar C, Muehlboeck JS, Simmons A (In press) Regional
magnetic resonance imaging measures for multivariate analysis in Alzheimers
disease and mild cognitive impairment. Brain Topogr DOI: 10.1007/s10548012-0246-x.
48. Kim S (2011) ppcor: Partial and Semi-partial (Part) correlation. The
Comprehensive R Archive Network website. URL http://CRAN.R-project.
org/package=ppcor. Accessed 2012 August 17. R package version 0.9-38.
49. Warnes GR, Bolker B, Bonebakker L, Gentleman R, Huber W, et al. (2011)
gplots: Various R programming tools for plotting data. The Comprehensive R
Archive Network website. URL http://CRAN.R-project.org/package=gplots.
Accessed 2012 August 17. R package version 0.9-38.
50. Stevenson M, Nunes T, Sanchez J, Thornton R, Reiczigel J, et al. (2012) epiR:
An R package for the analysis of epidemiological data. The Comprehensive R
Archive Network website. URL http://CRAN.R-project.org/package=epiR.
Accessed 2012 August 17. R package version 0.9-38.
51. Lindeman RH, Merenda PF, Gold RZ (1980) Introduction to Bivariate and
Multivariate Analysis. Scott Foreman, Glenview, IL, 120127.
52. Gromping U (2006) Relative importance for linear regression in R: the package
relaimpo. J Stat Softw 17: 127.
53. Lovestone S, Francis P, Kloszewska I, Meccoci P, Simmons A, et al. (2009)
AddNeuroMed the European collaboration for the discovery of novel
biomarkers for Alzheimers disease. Ann NY Acad Sci 1180: 3646.
32. Mold M, Shrive A (2012) Serum amyloid P component accelerates the formation
and enhances the stability of amyloid fibrils in a physiologically significant
under-saturated solution of amyloid-Ab42. J Alzheimers Dis 29: 875881.
33. de la Monte SM, Wands JR (2005) Review of insulin and insulin-like growth
factor expression, signaling, and malfunction in the central nervous system:
relevance to Alzheimers disease. J Alzheimers Dis 7: 4561.
34. Zambrano A, Otth C, Mujica L, Concha II, Maccioni RB (2007) Interleukin-3
prevents neuronal death induced by amyloid peptide. BMC Neurosci 8: 82.
35. Greco SJ, Sarkar S, Johnston JM, Tezapsidis N (2009) Leptin regulates tau
phosphorylation and amyloid through AMPK in neuronal cells. Biochem Bioph
Res Co 380: 98104.
36. Szczepanik AM, Funes S, Petko W, Ringheim GE (2001) IL-4, IL-10 and IL-13
modulate Ab(1-42)-induced cytokine and chemokine production in primary
murine microglia and a human monocyte cell line. J Neuroimmunol 113: 4962.
37. Bell RD, Winkler EA, Singh I, Sagare AP, Deane R, et al. (2012) Apolipoprotein
E controls cerebrovascular integrity via cyclophilin A. Nature 485: 512516.
38. Kohavi R (1995) A study of cross-validation and bootstrap for accuracy
estimation and model selection. Proceedings of IJCAI95 2: 11371143.
39. Alzheimers Association (2004) Research consent for cognitively impaired adults:
recommendations for institutional review boards and investigators. Alzheimer
Dis Assoc Disord 18: 171175.
MRI and CSF biomarkers in normal, MCI, and AD subjects: diagnostic
discrimination and cognitive correlations. Neurology 73: 287297.
41. Trojanowski JQ, Vandeerstichele H, Korecka M, Clark CM, Aisen PS, et al.
(2010) Update on the biomarker core of the Alzheimers Disease Neuroimaging
Initiative subjects. Alzheimers Dement 6: 230238.
42. Jagust WJ, Landau SM, Shaw LM, Trojanowski JQ, Koeppe RA, et al. (2009)
Relationships between biomarkers in aging and dementia. Neurology 73: 1193.
43. R Development Core Team (2011) R: A Language and Environment for
Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria.
URL http://www.R-project.org. ISBN 3-900051-07-0.
106
10
Generative FDG-PET and MRI Model of Aging and

Disease Progression in Alzheimers Disease
Juergen Dukart1,2*, Ferath Kherif1, Karsten Mueller2, Stanislaw Adaszewski1, Matthias L. Schroeter2,3,4,5,
Richard S. J. Frackowiak1, Bogdan Draganski1,2, for the Alzheimers Disease Neuroimaging Initiative"
1 LREN, Departement des Neurosciences Cliniques, CHUV, Universite de Lausanne, Lausanne, Switzerland, 2 Max-Planck-Institute for Human Cognitive and Brain Sciences,
Leipzig, Germany, 3 Day Clinic of Cognitive Neurology, University of Leipzig, Leipzig, Germany, 4 LIFE - Leipzig Research Center for Civilization Diseases, University of
Leipzig, Leipzig, Germany, 5 Consortium for Frontotemporal Lobar Degeneration, Leipzig, Germany
Abstract
The failure of current strategies to provide an explanation for controversial findings on the pattern of pathophysiological
changes in Alzheimers Disease (AD) motivates the necessity to develop new integrative approaches based on multi-modal
neuroimaging data that captures various aspects of disease pathology. Previous studies using [18F]fluorodeoxyglucose
positron emission tomography (FDG-PET) and structural magnetic resonance imaging (sMRI) report controversial results
about time-line, spatial extent and magnitude of glucose hypometabolism and atrophy in AD that depend on clinical and
demographic characteristics of the studied populations. Here, we provide and validate at a group level a generative
anatomical model of glucose hypo-metabolism and atrophy progression in AD based on FDG-PET and sMRI data of 80
patients and 79 healthy controls to describe expected age and symptom severity related changes in AD relative to a
baseline provided by healthy aging. We demonstrate a high level of anatomical accuracy for both modalities yielding
strongly age- and symptom-severity- dependant glucose hypometabolism in temporal, parietal and precuneal regions and a
more extensive network of atrophy in hippocampal, temporal, parietal, occipital and posterior caudate regions. The model
suggests greater and more consistent changes in FDG-PET compared to sMRI at earlier and the inversion of this pattern at
more advanced AD stages. Our model describes, integrates and predicts characteristic patterns of AD related pathology,
uncontaminated by normal age effects, derived from multi-modal data. It further provides an integrative explanation for
findings suggesting a dissociation between early- and late-onset AD. The generative model offers a basis for further
development of individualized biomarkers allowing accurate early diagnosis and treatment evaluation.
Citation: Dukart J, Kherif F, Mueller K, Adaszewski S, Schroeter ML, et al. (2013) Generative FDG-PET and MRI Model of Aging and Disease Progression in
Alzheimers Disease. PLoS Comput Biol 9(4): e1002987. doi:10.1371/journal.pcbi.1002987
Editor: Olaf Sporns, Indiana University, United States of America
Received December 11, 2012; Accepted January 27, 2013; Published April 4, 2013
Copyright: 2013 Dukart et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
Funding: JD is supported by the Synapsy NCCR of the Swiss National Science Foundation (http://www.nccr-synapsy.ch/). BD is supported by the Swiss National
Science Foundation (grant 320030_135679 and NCCR Synapsy), Foundation Parkinson Switzerland, Foundation Synapsys, Novartis Foundation for medical
biological research and Deutsche Forschungsgemeinschaft (Kfo 247). FK is supported by the Velux Stiftung. MLS is supported by the German Federal Ministry of
Education and Research (BMBF; German FTLD consortium, www.ftld.de/) and LIFE - Leipzig Research Center for Civilization Diseases at the University of Leipzig
(http://life.uni-leipzig.de/). LIFE is funded by means of the European Union, by the European Regional Development Fund (ERFD) and by means of the Free State
of Saxony within the framework of the excellence initiative. Data collection and sharing for this project was funded by the Alzheimers Disease Neuroimaging
Initiative (ADNI, adni.loni.ucla.edu) (National Institutes of Health Grant U01 AG024904). ADNI is funded by the National Institute on Aging, the National Institute of
Biomedical Imaging and Bioengineering, and through generous contributions from the following: Abbott, AstraZeneca AB, Bayer Schering Pharma AG, BristolMyers Squibb, Eisai Global Clinical Development, Elan Corporation, Genentech, GE Healthcare, GlaxoSmithKline, Innogenetics, Johnson and Johnson, Eli Lilly and
Co., Medpace, Inc., Merck and Co., Inc., Novartis AG, Pfizer Inc, F. Hoffman-La Roche, Schering-Plough, Synarc, Inc., as well as non-profit partners the Alzheimers
Association and Alzheimers Drug Discovery Foundation, with participation from the U.S. Food and Drug Administration. Private sector contributions to ADNI are
facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and
Education, and the study is coordinated by the Alzheimers Disease Cooperative Study at the University of California, San Diego. ADNI data are disseminated by
the Laboratory for Neuro Imaging at the University of California, Los Angeles. This research was also supported by NIH grants P30 AG010129, K01 AG030514, and
the Dana Foundation. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
" Please see the acknowledgments for further details of the Alzheimers Disease Neuroimaging Initiative.
Introduction
of glucose hypometabolism and atrophy [1214]. The supposition

that age- and symptom severity-related variability are the main
cause for these discrepancies has motivated researchers to adopt
analytical strategies that split disease populations into subgroups
depending on age- or symptom-severity (e.g. early- and late-onset
AD) [1214].
The question that needs to be answered is whether age and
symptom severity indeed account for most of these discrepancies.
In other words, what are the relative contributions of age and
Neuroimaging studies using [18F]fluorodeoxyglucose positron

emission tomography (FDG-PET) and structural magnetic resonance imaging (sMRI) provide substantial evidence of high
sensitivity for early detection and progression assessment in
Alzheimers disease (AD) at a group and single subject level [1
11]. However, among these studies there are a number of
discordant results in terms of spatial characteristics and magnitude
PLOS Computational Biology | www.ploscompbiol.org
107
Generative Biomarker Model in Alzheimers Disease
neurodegenerative processes. A second methodological limitation

of previous studies is related to pre-processing of FDG-PET data.
These data are strongly affected by the underlying atrophy
pattern. A reduction in grey matter volume in a specific region
would therefore also lead to a reduction of the observed metabolic
signal due to increased contribution of other tissue types. This
effect is commonly known as partial volume effect (PVE) [21]. If
not accounted for, this effect strongly restricts the interpretation of
the observed FDG-PET signal due to a high susceptibility to the
underlying atrophy. A correction for this effect is therefore
necessary to enable a valid interpretation of the independent
contribution of both imaging modalities to detection of AD related
pathology.
To address these issues and questions we generate group level
anatomical models of pathophysiological changes observed in AD
using FDG-PET and sMRI data. In these models we account for
PVE and integrate disease-, age- and symptom severity-associated
changes in AD patients. We further dissociate them from healthy
aging related changes using a combination of voxel-based general
linear models (GLMs). We additionally assume that AD-induced
changes are added to changes observed in healthy aging. We use
the models to generate age- and symptom severityspecific wholebrain patterns of glucose hypometabolism and atrophy. To
validate the obtained model and to address the questions described
above, we contrapose the models predictions in terms of
anatomical plausibility to findings reported in previous studies
investigating age- and AD-related changes. Thereby, we aim to
provide at a group level an integrative explanation for the
controversial findings described above regarding spatial characteristics and magnitude of glucose hypometabolism and atrophy in
AD. We further make conclusions on the relative capability of
FDG-PET and sMRI to predict AD related pathology at a group
level.
We hypothesize, on the basis of the above considerations [12
14] that a generative model predicting age and symptom severity
contributions to disease pathology based on data from two
imaging modalities FDG-PET and sMRI would provide a
robust and accurate differential pattern of glucose hypometabolism and atrophy at different ages in AD patients. We expect
stronger changes in glucose metabolic compared to anatomical
data in earlier disease stages, in accordance with a recently
proposed model, which suggests that functional impairment
precedes structural changes in AD [22]. We also hypothesize that
patterns of brain atrophy associated with healthy aging would
overlap those associated with disease progression yet also show a
clearly distinguishable anatomical distribution pattern.
Author Summary
Establishing an accurate diagnosis of Alzheimers disease
has been a major challenge in the past decades. With an
increasing amount of studies aiming at detection and
validation of imaging biomarkers for this disease, many
apparently controversial findings have been reported over
the time. The failure of current strategies to provide a
consistent explanation for these differential findings
motivates the necessity to develop new integrative
approaches based on multi-modal data that capture
various aspects of disease pathology. In our study we
propose such a generative model providing a comprehensive approach towards integration of previously
published differential findings in early- and late-onset
AD. We believe that our analytical strategy not only
provides the link between imaging biomarkers and clinical
phenotype considering the effects of aging, but could also
lead to new areas of research in terms of creation of new,
individualized biomarkers for a more accurate diagnosis of
Alzheimers disease.
108
disease to the anatomical patterns of abnormality of structure and

function? Investigation of these relationships may also provide
clues to another long-time controversy can the observed
differences between young and aged AD patients be regarded as
a continuum or is there a clear separation into two cohorts
dependent on separate pathological mechanisms?
Recent studies have suggested that AD related brain changes
may be similar to those associated with healthy aging. If so, this
could explain age overestimation determined from sMR images
from AD patients [15] and inaccuracies obtained with automated
classification-based computer diagnostics in the eldest healthy
controls and youngest AD patients [11]. However, these effects
can also be explained in a more simple way by the applied
methodology. Both AD and healthy aging have been linked to a
decrease in grey matter (GM) volume in extensive networks
covering substantial parts of the brain [1618]. Partial overlaps
between these networks are therefore very likely and also occur in
some key regions for AD including the hippocampus and parietal
cortices. A statistical classification/prediction model trained either
for prediction of age or AD would therefore consider these regions
as important for separation/prediction. An older healthy control
would therefore have a higher probability to be misclassified as AD
when applying a classifier trained on younger AD subjects. In
contrast, age prediction in an AD patient using a model built on
healthy controls would result in an age overestimation in this
patient due to the partially overlapping network of reductions in
GM volume. Both predictions are fully in line with previous
findings [15,19].
Another controversially discussed issue is the relative capability
and sensitivity of FDG-PET and sMRI to detect AD related
pathology. Recent studies provided evidence for the superiority of
each of the two imaging modalities as compared to the other to
detect AD related pathology [10,20]. However, none of these
findings can be interpreted without serious methodical considerations. Both studies restricted their analyses to univariate regionof-interest statistics to determine the power of each modality to
discriminate between AD patients and control subjects. Although
this approach provides a good estimation of the differential
pathology in the region mostly affected by the disease, it does not
at all reflect the whole pattern of AD related pathology over the
whole brain. Yet, exactly this whole-brain pattern is crucial for
detection and differentiation of AD from healthy aging and other
Methods
Subjects
To derive a generative model of age and symptom severity
related changes, we extracted from the Alzheimers disease
Neuroimaging Initiative (ADNI) database (www.adni-info.org)
sMRI and FDG-PET data from multiple centres of 80 patients
with a clinical diagnosis of AD and 79 healthy controls (Table 1). A
full list of subject and scan IDs used in this study is provided at the
following location: http://www.unil.ch/webdav/site/lren/shared/
Juergen/Overview_patandcon_MRIandPETdate.xlsx. For all subjects, follow-up evaluations were available for up to 5 years after
initial examination. Control subjects and AD patient groups were
matched for gender and age. A diagnosis of AD was based on
NINCDS/ARDRA criteria [23]. Exclusion criteria for the ADNI
data included the presence of any significant neurological disease
other than AD, history of head trauma followed by persistent
2

FDG-PET data
neurological deficits or structural brain abnormalities, psychotic

features, agitation or behavioural problems within the previous
three months or a history of alcohol or substance abuse. For most
subjects, multiple follow-up FDG-PET and sMRI scans were
available. Only data from the first examination date were used for
analysis. The study was conducted according to the Declaration of
Helsinki. Written informed consent was obtained from all
participants before protocol-specific procedures were performed.
Data used in the preparation of this article were obtained from the
Alzheimers Disease Neuroimaging Initiative (ADNI) database
(adni.loni.ucla.edu).
The ADNI was launched in 2003 by the National Institute on
Aging (NIA), the National Institute of Biomedical Imaging and
Bioengineering (NIBIB), the Food and Drug Administration
(FDA), private pharmaceutical companies and non-profit organizations, as a $60 million, 5- year public-private partnership. The
primary goal of ADNI has been to test whether serial magnetic
resonance imaging (MRI), positron emission tomography (PET),
other biological markers, and clinical and neuropsychological
assessment can be combined to measure the progression of mild
cognitive impairment (MCI) and early Alzheimers disease (AD).
Determination of sensitive and specific markers of very early AD
progression is intended to aid researchers and clinicians to develop
new treatments and monitor their effectiveness, as well as lessen
the time and cost of clinical trials. The Principal Investigator of
this initiative is Michael W. Weiner, MD, VA Medical Center and
University of California San Francisco. ADNI is the result of
efforts of many co-investigators from a broad range of academic
institutions and private corporations, and subjects have been
recruited from over 50 sites across the U.S. and Canada. The
initial goal of ADNI was to recruit 800 adults, ages 55 to 90, to
participate in the research, approximately 200 cognitively normal
older individuals to be followed for 3 years, 400 people with MCI
to be followed for 3 years and 200 people with early AD to be
followed for 2 years. For up-to-date information, see www.adniinfo.org.
We analysed FDG-PET data for subjects who also underwent

sMRI scans. FDG-PET data were acquired with different PETscanner types according to one of three different protocols: 1)
dynamic: a 30 min six-frame acquisition (6 five-minute frames),
with scanning from 30 to 60 min post FDG injection; 2) static: a
single-frame, 30 min acquisition with scanning 3060 min post
injection; and 3) quantitative: a 60 min dynamic protocol
consisting of 33 frames, with scanning beginning at injection and
continuing for 60 min. The majority of the scans in the ADNI
study were acquired with the first acquisition protocol. Images
further differed in resolution, orientation, voxel and image
dimensions and count statistics. The frames from 30 to 60 minutes
post injection were spatially realigned to minimize inter-frame
motion artefacts and a mean image of these frames was calculated
for each subject. These mean images were used for further
analysis.
Image pre-processing
All data processing steps were carried out using the SPM5
software package (Statistical Parametric Mapping software:
http://www.fil.ion.ucl.ac.uk/spm/) implemented in Matlab 7.7
(MathWorks Inc., Sherborn, MA). The same pre-processing
algorithm was used for sMRI and FDG-PET data, as described
elsewhere [11]. This procedure includes co-registration and
interpolation of both FDG-PET and sMR images to an isotropic
resolution of 16161 mm3, bias correction for inhomogeneity
artefacts for sMRI data, segmentation of sMRI data into different
tissue classes (only the grey matter tissue class is used for further
analyses), and masking of non-GM voxels in FDG-PET data. PVE
correction using the modified Muller-Gartner method [21,24] was
in the PVElab software package [25] that is compatible with
SPM5 only. This procedure uses the segmented sMR images to
account for PVE and for potential atrophy effects in FDG-PET.
DARTEL (Diffeomorphic Anatomical Registration using Exponentiated Lie algebra) based on grey matter tissue probability
maps was used for spatial normalization of data to an average size
template created from all study participants [26]. Structural MR
images were additionally modulated to preserve the total amount
of signal from each region. The same deformation matrices used to
normalise sMRI scans to a template were used to co-register the
FDG-PET images. After spatial normalization anatomical regions
of all subjects were located at same location in the images.
Smoothing with a Gaussian kernel of 12 mm FWHM (full width at
half maximum) accounted for minor misalignment errors.. FDGPET data were intensity normalized to cerebellar mean [27] and
masked to avoid big edge effects. The cerebellar region was chosen
for intensity normalization of FDG-PET as it has been shown to
be a region of choice for intensity normalization which is
unaffected in healthy aging and early stages of AD when
correcting for PVE caused by atrophy [2729].
sMRI data
The sMRI dataset included standard T1-weighted images
obtained with different scanner types using a 3D MP-RAGE
(magnetization-prepared 180 degrees radio-frequency pulses and
rapid gradient-echo) sequence varying in TR and TE (repetition
and echo time) with an in-plane resolution of 1.2561.25 mm and
1.2 mm slice thickness acquired at 1.5T magnetic field strength.
All raw data were pre-processed to correct for distortion and B1
non-uniformity as described on the ADNI webpage (http://www.
loni.ucla.edu/ADNI/Data/ADNI_Data.shtml).
Table 1. Subject group characteristics.
FDG-PET and sMRI models of healthy aging and AD

Controls
AD
T-test (df,t,p)
Number
79
80
Male/Female
41/38
40/40
Age (years)
75.864.9
75.767.0
157,0.1,.887
Age range (years)
6287
5588
MMSE (score)
28.761.6
23.662.2
157,16.6,,.001
All statistical analyses were also carried out using the SPM5
software package and Matlab 7.7. The effect of aging in healthy
control subjects was estimated separately for FDG-PET and sMRI
with voxel-wise linear regressions. To obtain the healthy aging
component of our generative model we used the beta coefficients
of aging in healthy controls to simulate voxel-wise changes in both
imaging modalities for the age range 50 to 80 years (Figure 1a).
The estimated values at 50 years were used as a 100% baseline.
Estimated age-related changes for the whole age range for both
FDG-PET and sMRI were expressed as percent decreases relative
to this baseline.
109
Mean 6 standard deviation. AD Alzheimers disease, con converters, MMSE Mini

Mental State Examination, noncon non-converters.
doi:10.1371/journal.pcbi.1002987.t001
Figure 1. Schematic representation of voxel-wise age- and symptom severity- related models. a) Schematic representation of agerelated changes in one voxel (in %) considering GM volume at the age of 50 years as baseline. b) Schematic representation of changes related to
healthy aging (black line) and age-related differences in AD (red line) in one voxel. Intersection age (dotted line) represents the age at which healthy
aging in this voxel becomes similar to changes observed in AD. The hinge in the red line (aging in AD) at the intersection point indicates that after the
intersection age, according to our assumption of the additive impact of AD related processes to healthy aging, the healthy aging model would apply
in AD patients as no pathological processes in terms of atrophy or glucose hypometabolism are longer observable after this time point. c) Decrease
(in %) in GM volume observed in an exemplary voxel in AD depending on the constellation of age and symptom severity (MMSE) relative to the
baseline provided by healthy aging (violet line). AD Alzheimers disease, GM grey matter, MMSE Mini Mental State Examination.
doi:10.1371/journal.pcbi.1002987.g001
110
To dissociate healthy aging- and AD-related changes, the

variance in glucose utilization and GM atrophy explained by
healthy aging was removed by voxel-wise linear regressions from
all imaging data used for further models (control subjects and AD
patients for both FDG-PET and sMRI; [19]). GLMs with age and
symptom severity (as measured by MMSE, [30]) as regressors were
built for the AD group separately for both FDG-PET and MRI.
To model the interaction of both factors the product of age and
symptom severity was also included. The inclusion of the
interaction between age and symptom severity accounts for any
differential disease progression associated with age in the AD
group. In this model we took the control group mean glucose

metabolism and GM volume at each voxel as a baseline. In an
effort to define the specific variance attributable to AD explained
by aging and symptom severity we removed variance from this
baseline that was explained by these factors and their interaction
(Figure 1c).
Further, we determined at what age the changes related to
normal aging become similar to those found in AD. For this
purpose, a third GLM, using age as the only factor, was calculated
in the AD cohort for both FDG-PET and sMRI data. The ages at
which the separate regressions for normal and AD associated
4

aging intersect (referred to as intersection ages) were calculated on

a voxel-wise basis (Figure 1b). To calculate these voxel-wise
intersection ages xage the linear regression equations for healthy
aging and aging in AD (yCon ~xage :bCon zcCon and
yAD ~xage :bAD zcAD , where yCon and yAD are the predicted
voxel-wise grey matter volumes/glucose metabolism, cCon and cAD
the corresponding intercepts and bCon and bAD the slopes of
regression lines in controls and AD respectively) were set equal and
resolved for xage . The intersection age xage is then given by:
xage ~
are restricted to calcarine gyri, insulae, anterior cingulate cortex,

superior temporal sulcus and posterior hippocampus.
The equivalent FDG-PET based model demonstrates a specific
pattern of age-related metabolic changes with an age related
decrease in glucose utilization of more than 10% per decade in
bilateral parietal, occipital, sensorimotor, premotor, dorsolateral
prefrontal and anterior insular cortices. Additionally, we see a
major reduction in glucose metabolism in bilateral posterior
putamina and in the left dorsal caudate nucleus (Figure 1b).
Age- and symptom severity- related changes
cCon {cAD
:
bAD {bCon
We report a negative relationship between symptom severity

and both metabolism and GM volume (Figure 3) and (Figure 4).
Lesser reductions in glucose utilization and GM volume are
associated with greater age in AD patients.
Symptom severity related GM volume changes at age 60 years
were detected throughout the brain with greatest degrees of
atrophy seen in bilateral parietal, temporal, occipital, dorsolateral
prefrontal, posterior cingulate and premotor cortices, the precuneus, dorsal caudate nucleus, amygdala and hippocampus. At 80
years of age, the greatest, bilateral, symptom severity related
atrophy was found in parietal, temporal, occipital, primary
sensorimotor and dorsolateral prefrontal cortices, the hippocampus and thalamus.
We show symptom severity- related glucose metabolism
reductions bilaterally in posterior temporal, parietal, lateral
occipital, dorsolateral prefrontal and premotor cortices and in
the precuneus. Symptom severity related hypometabolism is less
extensive at higher than lower ages.
In general, regional decreases in metabolism and grey matter
volumes relative to healthy aging are significantly more pronounced in the lower compared to higher age range in AD. In the
model, we observe substantial age-dependant differences in terms
of hypometabolism and atrophy in AD patients compared to a
healthy aging baseline even at an MMSE score of 30. At the age of
60 years these differences are bilaterally restricted to inferior
frontal gyrus, premotor regions, inferior and medial temporal
gyrus, cerebellum, rectal gyrus and to left parietal regions. Regions
showing an initial difference in this age range do not correspond
well to the anatomical pattern observed in later symptom severity
stages and remain rather less affected compared to other regions.
Initial differences observed at the age of 80 years are located in
bilateral parietal, bilateral hippocampal and left sensorimotor
regions and in both caudate nuclei.
We observe a more consistent anatomical pattern of initial
differences in hypometabolism at a MMSE score of 30. For the
whole age range of 6080 years, initial glucose hypometabolism is
observed in bilateral parietal, inferior temporal and posterior
cingulate cortices, posterior thalamus and the precuneus. Additionally, at 80 years of age we demonstrate significant differences
in bilateral primary sensorimotor and premotor regions and in the
anterior temporal lobes. All regions showing initial glucose
hypometabolism, except for the posterior thalamus, also show
the steepest symptom severity-related metabolic decline. Regions
hypometabolic only at age 80 show no specific symptom severityrelated decline.
Generally, an early intersection age indicates that a region is

relatively spared by AD, if healthy aging affects AD patients
similarly to controls. A higher intersection age suggests that the
region is predominantly affected by disease-related processes.
All three voxel-based models are based on a previously validated
assumption of linearity between healthy aging- and AD symptom
severity- related changes with FDG-PET and sMRI data. [16,31].
To exclude a more complex e.g., quadratic relationship between
age or symptom severity in the AD cohort we calculated further
regression analyses after removing variance explained by healthy
aging. These comprised a) including only linear relationships for
both age and symptom severity; b) additionally modelling
quadratic relationships with age; and c) additionally modelling a
quadratic relationship for symptom severity. Gender was included
as a covariate for both imaging modalities and total intracranial
volume was additionally included as a covariate for the sMRI data.
A significance threshold of 0.001 uncorrected at voxel-level and
0.05 family-wise error (FWE)-corrected for non-stationarity of
smoothness at cluster level was used for statistical analyses [32].
Statistical analysis of behavioural data

Group comparisons of AD patients and control subjects for age
and symptom severity were carried out using T-tests with a
significance threshold of p,.05. Group differences regarding
gender were evaluated using a chi-square test for independent
samples. The statistical analyses were performed with SPSS 17.0
(http://www.spss.com/statistics/). A Pearsons correlation coefficient was calculated (p,.05) in the AD group to investigate the
relationship between age and symptom severity.
Results
Subject demographic characteristics
AD patients and control subjects did not differ in age
[t(157) = 0.1;p = .887]. As expected MMSE differed significantly
between the groups [t(157) = 16.6;p,.001]. The comparison of
AD patients and control subjects in relation to sex showed no
statistical differences [x2(1) = 0.06;p = .811].
Differential pattern of changes related to healthy aging

in FDG-PET and sMRI
The generative model for healthy aging based on sMRI reveals
a widespread pattern of grey matter volume reductions sparing
only bilateral dorsal primary sensorimotor regions, brainstem,
lateral thalamus and the dorsal part of caudate nucleus (Figure 2a).
We observe the greatest reductions in GM volume, of more than
10% per decade, in right superior parietal lobule, superior and
inferior frontal gyrus, inferior frontal sulcus, primary auditory
cortex, pars triangularis, and anterior hippocampus. Left-hemispheric reductions are observed in the premotor cortex, and in
superior and middle frontal gyri. Bilateral GM volume reductions
Dissociating healthy aging from Alzheimers disease

To dissociate brain changes related to healthy aging from AD
pathology at specific ages we computed the intersection age of
models for healthy aging and aging in AD at the voxel level
(Figure 5). With sMRI, we see the highest intersection ages
bilaterally in hippocampus, anterior and posterior thalamus,
posterior and midcingulate, parietal, temporal, cerebellar, pre- 111
5
112

Figure 2. Healthy aging related changes observed in MRI (a) and FDG-PET (b) considering the expression of GM volume and
glucose metabolism at the age of 50 years as baseline. FDG-PET [18F]fluorodeoxyglucose positron emission tomography, GM grey matter, MRI
structural magnetic resonance imaging.
frontal and premotor regions. With FDG-PET, regions with the

highest intersection ages are bilaterally restricted to precuneus,
cerebellum, anterior and posterior cingulate, posterior parietal,
temporal, lateral occipital, primary motor, premotor and prefrontal cortices and the left dorsal caudate nucleus.
severity in younger compared to older AD patients. There were no

significant negative correlations. Further, we found significant
bilateral positive correlations with symptom severity in temporal,
lateral and medial prefrontal, inferior parietal, occipital regions
and in thalamus as well as the left premotor cortex. Again no
negative correlations were observed.
With FDG-PET we found a significant positive correlation
between age and glucose metabolism in bilateral temporal and
parietal regions. There were no other significant correlations with
age or symptom severity.
Inclusion of a quadratic relationship with age or MMSE into the
models revealed significant positive correlations of glucose metabolism with a quadratic age coefficient only in left dorsal parietal
cortex (Figure 5b). No other significant correlations were observed.
Linear relationship between age, symptom severity and

imaging results
With sMRI, the linear regression model describing the
relationship between age, symptom severity and structural changes
in AD showed a significant correlation for each factor with
atrophy (Figure 6a). After removing the variance attributable to
healthy aging, age correlates positively with GM volume in AD in
right premotor and bilaterally in parietal, temporal, occipital and
medial and lateral prefrontal regions. Additional significant
bilateral positive correlations were observed in anterior cingulate
cortex and in the cerebellum. These counter-intuitive positive
correlations can be interpreted as reflecting the additional agespecific atrophy needed to produce a similar degree of symptom
Discussion
In this study we demonstrate that a generative model captures
the anatomical and metabolic features associated with AD and
Figure 3. A linear model of age and MMSE related changes observed in AD in FDG-PET considering the healthy control group as
baseline. AD Alzheimers disease, FDG-PET [18F]fluorodeoxyglucose positron emission tomography, MMSE Mini Mental State Examination.
113
Figure 4. A linear model of age and MMSE related changes observed in AD in MRI considering the healthy control group as
baseline. AD Alzheimers disease, MMSE Mini Mental State Examination, MRI structural magnetic resonance imaging.
114
hypometabolism and atrophy was observed between a network of

areas affected by healthy aging and that identified in AD. To
interpret this overlap it is important to note that all changes related
to age and symptom severity in AD we report here were calculated
after removing the variance explained by healthy aging. That
means that all age-related differences can be interpreted as an addon required to normal age-related changes to induce a predefined
symptom severity at the corresponding age. In previous research
comparing differently aged AD groups with age-matched healthy
control subjects a split of AD into different subgroups e.g., earlyand late-onset AD, was suggested [12]. In contrast, the results of
our study indicate that the anatomically different qualitative and
quantitative patterns observed in AD at different ages may be
explicable by regionally inhomogeneous, age-related, baseline
changes in healthy controls. This notion is supported by the
observation of a similar anatomical pattern of structural and
metabolic changes across the studied age range in AD patients.
Overall, our results suggest a more integrative view of AD
indicating that previously reported differences between early- and
late-onset AD patients can be explained as an interaction of AD
pathology with changes due to healthy aging. More simple,
assuming that AD affects parietal and temporal regions, a strong
healthy aging related decrease in glucose metabolism e.g. in the
parietal cortex with at the same time relatively preserved
healthy aging in the age range 60 to 80 years accurately and

robustly. The model differentiates between the effects of aging and
symptom severity in AD patients. It also provides a means to test
for interactions between them, as exemplified here with disease
progression and patient age. The age-dependant differential
sensitivity of structural and metabolic scanning for the detection
of AD pathology we demonstrate emphasises the ability of our
generative model to infer expected age- and symptom severityspecific changes recorded with both imaging modalities.
Our studys three main findings support and extend observations made previously. Firstly, the spatial location of changes
related to different ages and symptom severities in AD was mostly
consistent with regions reported in the literature [3335].
However we also identified a more widespread network of regions
showing age related metabolic decreases and atrophy whilst
controlling for potential PVE related changes in FDG-PET. The
AD model at lower ages was substantially more extensive in terms
of decreases in glucose metabolism and GM volume than at
greater ages. This finding is in accordance with previous research
[1214].
Our second result is that in general, the magnitude of healthy
aging related imaging changes between 50 and 80 years is
comparable to that of changes associated with an increase in AD
symptom severity at any age. Substantial regional overlap in both

Figure 5. Voxel-wise intersections of healthy aging and changes observed in AD in MRI (a) and FDG-PET (b). Colour bars represent
the intersection age. AD Alzheimers disease, FDG-PET [18F]fluorodeoxyglucose positron emission tomography, MRI structural magnetic resonance
imaging.
metabolism in the temporal region would lead to an observation of

both regions when comparing younger patients to younger
controls. Yet, only the temporal region would be detected when
comparing older patients to older controls.
A third result is in the dissociation of age- and disease- related
processes inferred from the intersection ages of healthy aging and
AD models. While in previous research some studies successfully
applied whole-brain approaches to discriminate dementia patients
from control subjects [6,11,36], sparse solutions based on highly
discriminative features have been used in most [4,3740].
However, there is a problem with feature selection that has been
neglected till now. This is the assumption of feature reliability,
independent of the expression of possible confounding variables.
We decided to investigate this assumption by testing whether a
correlation of AD symptom severity with imaging data was also
affected by the processes of normal aging as defined in a healthy
population. We did this with our described method of model
intersections. Our results suggest that features like local atrophy or
glucose hypometabolism that discriminate between AD patients
and control subjects at the age of 60 are not necessarily observed at
age 80. This non-stationarity of features is mainly explained by
atrophy and hypometabolism related to healthy aging, which
affect some regions similarly to AD. We previously demonstrated
that accounting for changes related to healthy aging improves AD
detection with support vector machine classification [19]. This
study extends that approach and suggests that diagnostic classifier
algorithms, like machine learning techniques, when applied to the
general population, need to take potential interactions of imaging
features with demographic and clinical factors into account.
Related to this, our intersection model also provides evidence that
the pathological pattern observed in AD in some regions is clearly
distinguishable from healthy aging even at very advanced age.
The obtained models could be used to improve early AD
detection e.g. by training automated classifiers on age- and
symptom severityspecific pathological AD patterns (features)

extracted by thresholding the obtained AD model. They could also
be applied directly in clinical assessment by evaluating the
similarity of observed pathology in any individual to age- and
symptom severityspecific patterns generated using the AD model.
Thereby, to enable individual assessment, percent signal difference
maps could be calculated between each subjects imaging data and
imaging data generated using the healthy aging model. However,
both of these approaches require careful evaluation in future
studies prior to clinical application.
Nonetheless, a valid interpretation of our results needs to
consider the effects of several other assumptions and limitations.
First, the sensitivity of the generative model for detecting and
predicting AD pathology depends on the accuracy of the model of
healthy aging. The age- related, widespread patterns of brain
atrophy and hypometabolism we find are consistent with previous
findings [16,31,4143]. The most prominent changes of glucose
hypometabolism we observed include large parts of an occipitalparietal-frontal network. The only non-linear (quadratic) relationship between age and metabolic changes we found in an
anatomically highly restricted area in the left dorsal parietal
cortex. The absence of more complex relationships between these
parameters indicates that linear models are very probably a
sufficient approximation of the underlying structural and metabolic processes.
A major advantage of our approach is the generalizability of our
model to any constellation of age and symptom severity. A further
difference to conventional approaches is that the method we
propose relies on a voxel-wise group mean, ignoring the variance
and so allowing detection of quite minor group differences. This
type of analysis, though more flexible than conventional statistics,
nevertheless requires caution in the interpretation of results. Minor
differences between groups of healthy subjects and AD patients
could still be due to random effects unrelated to AD. However, as
115
Figure 6. Positive linear (a) or quadratic (b) relationship observed between age and MMSE in MRI (blue) and FDG-PET (red) in AD
patients after removing the variance explained by heathy aging. Only clusters are shown exceeding a significance threshold of p = 0.001
uncorrected on voxel level and p = 0.05 FWE-corrected on cluster level. AD Alzheimers disease, FDG-PET [18F]fluorodeoxyglucose positron emission
tomography, MRI structural magnetic resonance imaging.
we model disease progression (albeit based on cross-sectional data)

we would expect the magnitude of differences to increase with
higher symptom severity. This assumption suggests that local
differences between AD patients and control subjects that are not
so correlated with it are more likely to be due to random artefacts
than disease related pathology.
For all three models used in our study, we made the assumption
that healthy aging affects AD patients in the same way as healthy
subjects in terms of hypometabolic and atrophic changes, with AD
pathology additive to changes associated with healthy aging. This
assumption is in line with current views that AD as a pathological
process is unrelated to healthy aging.
A further issue of interpretation of interactions is the recognised
inaccuracy of clinical diagnosis of AD, which may differ in
younger vs. older or in mildly vs. severely impaired patients. Any
bias towards an alternative diagnosis or co-diagnosis with another
dementing condition could lead to a different anatomical or
hypometabolic pattern. Such a differential pattern of glucose
hypometabolism was seen, for example, in sensorimotor regions
only in old AD patients with low symptom severity.
Most of these problems are also common to standard statistical
methods of evaluation of group differences. In group statistics one
often uses a high significance threshold to avoid false positive
results thus minimising their effect on between group differenti-
ations. By contrast, our approach also provides an opportunity to

evaluate the impact of possible confounding effects, such as age in
this case, on the discrimination between AD patients and control
subjects.
Text S1 Acknowledgment list for ADNI publications.
(PDF)
Acknowledgments
We thank Jessica Peters for preparing parts of the data used in this project.
investigators can be found in Text S1.
Conceived and designed the experiments: JD MLS RSJF BD. Performed
the experiments: JD. Analyzed the data: JD. Contributed reagents/
materials/analysis tools: JD FK KM SA. Wrote the paper: JD FK KM SA
MLS RSJF BD.
116
10

References
1. Mielke R, Pietrzyk U, Jacobs A, Fink GR, Ichimiya A, et al. (1994) HMPAO
SPET and FDG PET in Alzheimers disease and vascular dementia: comparison
of perfusion and metabolic pattern. Eur J Nucl Med 21: 10521060.
2. Laakso MP, Soininen H, Partanen K, Helkala EL, Hartikainen P, et al. (1995)
Volumes of hippocampus, amygdala and frontal lobes in the MRI-based
diagnosis of early Alzheimers disease: correlation with memory functions.
J Neural Transm Park Dis Dement Sect 9: 7386.
3. Silverman DH, Gambhir SS, Huang HW, Schwimmer J, Kim S, et al. (2002)
Evaluating early dementia with and without assessment of regional cerebral
metabolism by PET: a comparison of predicted costs and benefits. J Nucl Med
43: 253266.
5. Habeck C, Foster NL, Perneczky R, Kurz A, Alexopoulos P, et al. (2008)
Multivariate and univariate neuroimaging biomarkers of Alzheimers disease.
Automatic classification of MR scans in Alzheimers disease. Brain 131: 681
689.
7. Sadeghi N, Foster NL, Wang AY, Minoshima S, Lieberman AP, et al. (2008)
Automatic classification of Alzheimers disease vs. Frontotemporal dementia: A
spatial decision tree approach with FDG-PET. In: 5th IEEE International
Symposium on Biomedical Imaging: From Nano to Macro; 1417 May 2008;
Paris, France. Vols 14: 408411.
8. Haense C, Herholz K, Jagust WJ, Heiss WD (2009) Performance of FDG PET
for detection of Alzheimers disease in two independent multicentre samples
(NEST-DD and ADNI). Dement Geriatr Cogn Disord 28: 259266.
9. Schroeter ML, Stein T, Maslowski N, Neumann J (2009) Neural correlates of
Alzheimers disease and mild cognitive impairment: a systematic and
quantitative meta-analysis involving 1351 patients. Neuroimage 47: 11961206.
10. Morinaga A, Ono K, Ikeda T, Ikeda Y, Shima K, et al. (2010) A comparison of
the diagnostic sensitivity of MRI, CBF-SPECT, FDG-PET and cerebrospinal
fluid biomarkers for detecting Alzheimers disease in a memory clinic. Dement
Geriatr Cogn Disord 30: 285292.
11. Dukart J, Mueller K, Horstmann A, Barthel H, Moller HE, et al. (2011)
Combined Evaluation of FDG-PET and MRI Improves Detection and
Differentiation of Dementia. PLoS One 6: e18111.
12. Sakamoto S, Ishii K, Sasaki M, Hosaka K, Mori T, et al. (2002) Differences in
cerebral metabolic impairment between early and late onset types of Alzheimers
disease. J Neurol Sci 200: 2732.
13. Ishii K, Kawachi T, Sasaki H, Kono AK, Fukuda T, et al. (2005) Voxel-based
morphometric comparison between early- and late-onset mild Alzheimers
disease and assessment of diagnostic performance of z score images. AJNR
Am J Neuroradiol 26: 333340.
14. Frisoni GB, Pievani M, Testa C, Sabattoli F, Bresciani L, et al. (2007) The
topography of grey matter involvement in early and late onset Alzheimers
15. Franke K, Ziegler G, Kloppel S, Gaser C (2010) Estimating the age of healthy
subjects from T1-weighted MRI scans using kernel methods: exploring the
influence of various parameters. Neuroimage 50: 883892.
16. Good CD, Johnsrude IS, Ashburner J, Henson RN, Friston KJ, et al. (2001) A
voxel-based morphometric study of ageing in 465 normal adult human brains.
17. Baron JC, Chetelat G, Desgranges B, Perchey G, Landeau B, et al. (2001) In
vivo mapping of gray matter loss with voxel-based morphometry in mild
18. Teipel SJ, Wegrzyn M, Meindl T, Frisoni G, Bokde AL, et al. (2012) Anatomical
MRI and DTI in the Diagnosis of Alzheimers Disease: A European Multicenter
Study. J Alzheimers Dis 31: S3347.
19. Dukart J, Schroeter ML, Mueller K, Initiative TAsDN (2011) Age Correction in
Dementia - Matching to a Healthy Brain. PLoS One 6: e22193.
20. Karow DS, McEvoy LK, Fennema-Notestine C, Hagler DJ, Jennings RG, et al.
(2010) Relative capability of MR imaging and FDG PET to depict changes
associated with prodromal and early Alzheimer disease. Radiology 256: 932
942.
21. Muller-Gartner HW, Links JM, Prince JL, Bryan RN, McVeigh E, et al. (1992)
Measurement of radiotracer concentration in brain gray matter using positron
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
emission tomography: MRI-based correction for partial volume effects. J Cereb

Blood Flow Metab 12: 571583.
Jack CR, Knopman DS, Jagust WJ, Shaw LM, Aisen PS, et al. (2010)
cascade. Lancet Neurol 9: 119128.
McKhann G, Drachman D, Folstein M, Katzman R, Price D, et al. (1984)
Rousset OG, Ma Y, Evans AC (1998) Correction for partial volume effects in
PET: principle and validation. J Nucl Med 39: 904911.
Quarantelli M, Berkouk K, Prinster A, Landeau B, Svarer C, et al. (2004)
Integrated software for the analysis of brain PET/SPECT studies with partialvolume-effect correction. J Nucl Med 45: 192201.
Ashburner J (2007) A fast diffeomorphic image registration algorithm. Neuroimage 38: 95113.
Dukart J, Mueller K, Horstmann A, Vogt B, Frisch S, et al. (2010) Differential
effects of global and cerebellar normalization on detection and differentiation of
dementia in FDG-PET studies. Neuroimage 49: 14901495.
Yanase D, Matsunari I, Yajima K, Chen W, Fujikawa A, et al. (2005) Brain
FDG PET study of normal aging in Japanese: effect of atrophy correction.
Eur J Nucl Med Mol Imaging 32: 794805.
Salmon E, Collette F, Degueldre C, Lemaire C, Franck G (2000) Voxel-based
analysis of confounding effects of age and dementia severity on cerebral
metabolism in Alzheimers disease. Hum Brain Mapp 10: 3948.
Folstein MF, Folstein SE, McHugh PR (1975) Mini-mental state. A practical
Res 12: 189198.
Loessner A, Alavi A, Lewandrowski KU, Mozley D, Souder E, et al. (1995)
Regional cerebral function determined by FDG-PET in healthy volunteers:
normal patterns and changes with age. J Nucl Med 36: 11411149.
Hayasaka S, Phan KL, Liberzon I, Worsley KJ, Nichols TE (2004)
Nonstationary cluster-size inference with random field and permutation
methods. Neuroimage 22: 676687.
Willis MW, Ketter TA, Kimbrell TA, George MS, Herscovitch P, et al. (2002)
Age, sex and laterality effects on cerebral glucose metabolism in healthy adults.
Psychiatry Res 114: 2337.
Zuendorf G, Kerrouche N, Herholz K, Baron JC (2003) Efficient principal
component analysis for multivariate 3D voxel-based mapping of brain functional
imaging data sets as applied to FDG-PET and normal aging. Hum Brain Mapp
18: 1321.
Kalpouzos G, Chetelat G, Baron JC, Landeau B, Mevel K, et al. (2009) Voxelbased mapping of brain gray matter volume and glucose metabolism profiles in
normal aging. Neurobiol Aging 30: 112124.
Kloppel S, Stonnington CM, Barnes J, Chen F, Chu C, et al. (2008) Accuracy of
dementia diagnosis: a direct comparison between radiologists and a computerized method. Brain 131: 29692974.
Davatzikos C, Fan Y, Wu X, Shen D, Resnick SM (2008) Detection of
prodromal Alzheimers disease via pattern classification of magnetic resonance
imaging. Neurobiol Aging 29: 514523.
Fan Y, Resnick SM, Wu X, Davatzikos C (2008) Structural and functional
biomarkers of prodromal Alzheimers disease: a high-dimensional pattern
classification study. Neuroimage 41: 277285.
Mosconi L, Tsui WH, Herholz K, Pupi A, Drzezga A, et al. (2008) Multicenter
standardized 18F-FDG PET diagnosis of mild cognitive impairment, Alzheimers disease, and other dementias. J Nucl Med 49: 390398.
Chaves R, Ramirez J, Gorriz JM, Lopez M, Salas-Gonzalez D, et al. (2009)
SVM-based computer-aided diagnosis of the Alzheimers disease using t-test
NMSE feature selection with feature correlation weighting. Neurosci Lett 461:
293297.
Sowell ER, Thompson PM, Toga AW (2004) Mapping changes in the human
cortex throughout the span of life. Neuroscientist 10: 372392.
Smith CD, Chebrolu H, Wekstein DR, Schmitt FA, Markesbery WR (2007) Age
and gender effects on human brain anatomy: a voxel-based morphometric study
in healthy elderly. Neurobiol Aging 28: 10751087.
Giorgio A, Santelli L, Tomassini V, Bosnell R, Smith S, et al. (2010) Age-related
changes in grey and white matter structure throughout adulthood. Neuroimage
51: 943951.
117
11
Evaluating Alzheimers Disease Progression Using Rate

of Regional Hippocampal Atrophy
Edit Franko1,2,3*, Olivier Joly4,5, for the Alzheimers Disease Neuroimaging Initiative
1 INSERM U1075, Universite de Caen Basse-Normandie, Caen, France, 2 Department of Neurodegenerative Disease, Institute of Neurology, University College London,
London, United Kingdom, 3 National Prion Clinic, Institute of Neurology, University College London, London, United Kingdom, 4 NeuroSpin, Gif-sur-Yvette, France,
5 Institute of Neuroscience, Newcastle University, Newcastle upon Tyne, United Kingdom
Abstract
Alzheimers disease (AD) is characterized by neurofibrillary tangle and neuropil thread deposition, which ultimately results in
neuronal loss. A large number of magnetic resonance imaging studies have reported a smaller hippocampus in AD patients
as compared to healthy elderlies. Even though this difference is often interpreted as atrophy, it is only an indirect
measurement. A more direct way of measuring the atrophy is to use repeated MRIs within the same individual. Even though
several groups have used this appropriate approach, the pattern of hippocampal atrophy still remains unclear and difficult
to relate to underlying pathophysiology. Here, in this longitudinal study, we aimed to map hippocampal atrophy rates in
patients with AD, mild cognitive impairment (MCI) and elderly controls. Data consisted of two MRI scans for each subject.
The symmetric deformation field between the first and the second MRI was computed and mapped onto the threedimensional hippocampal surface. The pattern of atrophy rate was similar in all three groups, but the rate was significantly
higher in patients with AD than in control subjects. We also found higher atrophy rates in progressive MCI patients as
compared to stable MCI, particularly in the antero-lateral portion of the right hippocampus. Importantly, the regions
showing the highest atrophy rate correspond to those that were described to have the highest burden of tau deposition.
Our results show that local hippocampal atrophy rate is a reliable biomarker of disease stage and progression and could also
be considered as a method to objectively evaluate treatment effects.
Citation: Franko E, Joly O, for the Alzheimers Disease Neuroimaging Initiative (2013) Evaluating Alzheimers Disease Progression Using Rate of Regional
Hippocampal Atrophy. PLoS ONE 8(8): e71354. doi:10.1371/journal.pone.0071354
Editor: Karl Herholz, University of Manchester, United Kingdom
Received March 13, 2013; Accepted June 28, 2013; Published August 12, 2013
Copyright: 2013 Franko et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
Funding: The authors of this study have only been funded by the University of Caen and INSERM. Data collection and sharing for this project was funded by the
Alzheimers Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904). ADNI was funded by the National Institute on Aging, the
National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: Abbott, AstraZeneca AB, Bayer Schering
Pharma AG, Bristol-Myers Squibb, Eisai Global Clinical Development, Elan Corporation, Genentech, GE Healthcare, GlaxoSmithKilne, Innogenetics, Johnson and
Johnson, Eli Lilly and Co., Medpace, Inc., Merck and Co., Inc., Novartis AD, Pfizer Inc, F. Hoffman-La Roche, Schering-Plough, Synarc, Inc., as well as non-profit
partners the Alzheimers Association and Alzheimers Drug Discovery Foundation, with participation from the U.S. Food and Drug Administration. Private sector
contributions to ADNI are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California
Institute for Research and Education, and the study is coordinated by the Alzheimers Disease Cooperative Study at the University of California, San Diego. ADNI
data are disseminated by the Laboratory for Neuro Imaging at the University of California, Los Angeles. This research was also supported by NIH grants P30
AG010129, K01 AG030514, and the Dana Foundation. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of
the manuscript.
Introduction
(MRI) studies have indeed reported smaller volumes in AD

patients than in controls [810], which indirectly reflects faster
atrophy in AD. However, a more direct way of measuring the
atrophy is the use of repeated MRI scans within the same
individual. This can then be used to monitor the disease
progression in patients. More and more volumetric studies used
longitudinal dataset and reported a higher rate of hippocampal
volume loss in patients with AD than in elderly controls [1117]
However, global hippocampal volumetry is not always sensitive
enough to follow changes within a single population [18], which
may reflect conversion from healthy state or disease progression.
Therefore, recent studies focus on changes in hippocampal shape.
Examining the shape of the hippocampus gives not only more
sensitivity to follow the progression of the atrophy, but also allows
the evaluation of atrophy in the different parts of the hippocampus. Thompson et al. [19] and Morra et al. [20,21] reported that
atrophy indeed varies by hippocampal sub-region. Based on
differences in atrophy, they could distinguish healthy subjects from
Alzheimers disease (AD) is the most common form of dementia

in the elderly population [1]. In 2010, the estimated number of
patients with AD worldwide was 35.6 million, and this number is
predicted to triple by 2050 (World Alzheimer Report, 2010).
These numbers highlight a pressing need to develop diseasemodifying treatments. Existing treatments are only effective in the
early phase of the disease, and even then, their effect is highly
variable among patients [2].
At present, the clinical diagnosis of AD requires that the patient
has dementia [3], which is already associated with the widespread
deposition of amyloid plaques and neurofibrillary tangles in the
brain [4]. Indeed, amyloid and tau deposition in the entorhinal
cortex has been detected in clinically silent cases [5]. Even in mild
AD, these neuropathological changes cause neuronal loss in the
entorhinal cortex and hippocampus [6], and these changes result
in decreased volume [7]. Many magnetic resonance imaging
118
August 2013 | Volume 8 | Issue 8 | e71354

Hippocampal Atrophy Rate in Alzheimers Disease
3.9 ms, flip angle 8, and slice thickness 1.2 mm. In-plane
resolution differed slightly among subjects: 0.9460.94 mm,
1.2561.25 mm, or 1.361.3 mm.In this study, we used two scans
from each subject. MRI_acq1 was the baseline scan (first scan of
the subject), and MRI_acq2 was the last available scan (through
December 2011). All subjects included in the analyses had a
duration of at least 200 days between the first and last scans.
patients with AD and demonstrated correlation between atrophy

and cognitive decline. However, their atrophy maps differed
notably; Thompson et al. [19] described higher atrophy in the
supero-lateral side of the hippocampus in AD compared to
controls, whereas Morra et al. [20] found the main difference
between groups in the inferior hippocampal surface. Surprisingly,
the regions showing significant difference between groups did not
correspond to the regions with significant atrophy in AD. These
contradictory results ask for further research to find more precise
methods to measure the hippocampal deformation during the
progression of the disease.
In the present longitudinal study, we examined the hippocampal
atrophy rate and compared its topography among healthy people
and patients with MCI and AD to identify areas that can
distinguish between groups. We found similar patterns of atrophy
among the groups. Regional higher atrophy rate was found in AD
as compared to controls. These regions also distinguish patients
with stable MCI from those who eventually progressed to AD.
Finally, the atrophy pattern is in agreement with the known
regional anatomic specificity of tau deposition.
Hippocampal Segmentation
A template hippocampus (right hemisphere) was manually
segmented from the single-subject MNI-T1 template available in
SPM (www.fil.ion.ucl.ac.uk/spm). The segmentation followed the
description of Franko et al. [23] and Insausti and Amaral [24] and
contains the hippocampus proper (cornu ammonis (CA) 13
fields), the dentate gyrus and the subiculum. For guided extraction
of the subjects hippocampus in MRI_acq1, two sets of seven
points were defined in the template and in the MRI_acq1,
respectively. The locations of the points are similar to those used
by Pluta et al. [25]; three points were placed on coronal slices and
four points on sagittal slices. The first point on the coronal view
marks the appearance of the hippocampus inferior to the
amygdala, the second marks the most medial point of the
hippocampus at the level of the hippocampal-amygdaloid
transitional area, and the third point indicates the most posterior
part of the hippocampus [23]. On the sagittal view, we placed the
first point on the most lateral slice of the hippocampus, the second
and the third were placed 23 mm more medial on the anterior
and posterior borders of hippocampus, respectively, and the last
point was placed at the end of the intralimbic gyrus lateral to the
hippocampal fissure. Finally, to extract the 3D hippocampus from
MRI_acq1, the Iterative Closest Point (ICP) algorithm (www.vtk.
org) was used to compute the affine transformation between the
template and the subjects hippocampus (Fig. 1A).
Methods
Subjects
Data used in the preparation of this article were obtained from
the Alzheimers Disease Neuroimaging Initiative (ADNI) database
(adni.loni.ucla.edu), for up-to-date information, see www.adniinfo.org. The data were analysed anonymously, using publicly
available secondary data from the ADNI study, therefore no ethics
statement is required for this work.
From the ADNI database, a preliminary dataset containing AD
and normal subjects were downloaded first while developing the
method [22]. From this first dataset, we selected the subjects who
had at least 2 scans with at least 200 days apart. This resulted in
90 AD patients and 54 control subjects. Later we expanded the
dataset to have similar number of controls as patients and added
the MCI group. From the downloaded subjects, we included the
first about 90 subjects who fulfilled the inclusion criteria, namely
having at least two MRIs with at least 200 days apart. Few control
subjects, who turned to MCI or AD during the follow-up period,
were then removed from the statistical tests. The final number of
subjects included in this study were 85 healthy controls, 102
patients diagnosed with MCI and 90 patients with AD. An
overview of the subject groups is provided in Table 1, including
the total number of subjects, their age, sex, mini-mental state
examination (MMSE) and clinical dementia rating (CDR) scores.
The interscan intervals are also shown for each group. We further
divided the MCI group into progressive MCI (pMCI, n = 39),
containing subjects who converted to AD between the baseline
and the last scan (CDR.0.5), and stable MCI (sMCI, n = 44),
containing subjects with CDR scores of 0.5 at the time of the last
scan. The subjects, whose CDR score was not available at the time
of the last scan, were not included into this analysis. Almost all
patients with AD (88 out of 90) received medication (cholinesterase
inhibitors and NMDA-receptor antagonists), as did 78% of the
patients with MCI (80 out of 102) whereas none of the controls.
Measurements of Atrophy
Following the recommendation of Yushkevich et al. [26], we
performed an unbiased symmetric deformation field estimation to
measure the hippocampal atrophy rate. ANTS software (www.
picsl.upenn.edu/ANTS) was used to compute the symmetric
deformation field (SyN method) between MRI_acq1 and
MRI_acq2 using a normalised cross-correlation metric. The
displacement of each voxel in MRI_acq1 was written in the
WarpX, WarpY, and WarpZ images of the deformation fields.
The parameters of our deformation-based morphometry included
isotropic 2 mm (FWHM Gaussian kernel) image smoothing.
Finally, for each vertex of the hippocampal surface (MRI_acq1),
the dot product between the normal and the deformation field
defined the signed displacement of the vertex (in mm) (Fig. 1B).
The resulting values (divided by the duration between acquisition
MRI_acq1 and MRI_acq2) represent the atrophy rate in mm/
year. The texture was displayed on the 3D hippocampus in the
radiological convention with visualisation software (anatomist,
brainvisa.info/). The 3D mesh together with statistical maps is also
available: brainsenses.x10host.com/hc.htm. The hippocampal
volume was computed from the volumetric meshes. The technique
described here was partly published in abstract form with
preliminary results [22].
MRI Acquisitions
Statistical Analyses
Structural brain MRIs were acquired at multiple ADNI sites

using 1.5 Tesla MRI scanners manufactured by General Electric
Healthcare, Siemens Medical Solutions, and Philips Medical
Systems. Sagittal 3D MP-RAGE sequences were acquired using
standard ADNI protocols (adni.loni.ucla.edu). MRI acquisition
parameters were as follows: repetition time 89 ms, echo time
Statistical analyses were performed using the R software (www.

cran.r-project.org). The permutation test was used to assess the
significant atrophy rate within each group. Statistical maps are
displayed on the 3D hippocampi as -log10(p-value) for above
threshold. Group differences were assessed using the Wilcoxon 119
2
Table 1. Characteristics of subjects in each group.
Group
Age in years
MMSE
CDR
Interscan time in days
Control (n = 85) F:37 M:48
76 (5)
29.2 (1)
0 (0)
1217 (303)
MCI (n = 102) F:34 M:68
75 (7)
26.8 (1.8)
0.5 (0)
1030 (393)
AD (n = 90) F:42 M:48
75 (7)
23.3 (2.2)
0.83 (0.35)
652 (188)
Mean values are followed by standard deviations (SD). MMSE: mini-mental state examination; CDR: clinical dementia rating; F: female; M:male.
in controls in both hippocampi (right: W = 5390, p-value = 1.501e06; left: W = 4690, p-value = 0.004927). The volume loss was
significantly higher in patients with MCI than in controls in the
right hemisphere but not the left (right: W = 5273, p-value = 0.005484; left: W = 4719, p-value = 0.149 n.s.). Similarly,
the AD group showed significantly greater volume loss than the
MCI group in the right but not in the left hippocampus (right:
W = 5507,
p-value = 0.008535;
left:
W = 5190,
p-value = 0.05936 n.s.). No significant difference in volume change
was found between progressive and stable MCI (right: W = 1025,
p-value = 0.06436 n.s.; left: W = 996, p-value = 0.1048 n.s.). Collectively, the volumetric measurements revealed significant loss in
all groups and a larger loss in the AD and the MCI groups than in
controls, but they failed to show significant difference between
pMCI and sMCI.
signed-rank test, which does not require the normality of the data.
For both type of tests, significance level (p,0.05) was corrected for
multiple comparison (number of vertices) using Bonferroni
correction.
Hippocampal Subfields
Even though the hippocampal subfields are determined by
histology and their boundaries are not visible on MRI, the
subfields are often indicated on in vivo MRIs based solely on
visual cues derived from histological images. However, a more
reliable and unbiased subfield delineation can be achieved by
using an independent MRI atlas. Hence, to help localise significant
atrophy rates on the hippocampus, we projected the hippocampal
subfields as defined from the high-resolution atlas of the human
hippocampus [27] computed from post-mortem MRI at 9.4 Tesla
(www.nitrc.org/projects/pennhippoatlas) onto our template using
ICP algorithm. The borders of the subfields including CA1, CA2
3, the dentate gyrus, and the subiculum are illustrated as outlines
on the 3D surface of the right hippocampus in Fig. 2A.
Mapping the Rate of Atrophy

We computed averaged maps of hippocampal atrophy rates
(mm/year) for patients with AD (Fig. 2A), MCI (Fig. 2B), and for
controls (Fig. 2C). This mapping revealed a similar pattern of
atrophy rates in all groups. Nonetheless, the rate of atrophy was
the highest in the AD group (demonstrated by darker blue in
Fig. 2), particularly in CA1 and subiculum. To assess atrophy rate
significance, statistical maps were derived from a permutation test
within each of the three populations (Fig. 3 AC). Again, a similar
pattern was found in the three groups. Significant atrophy
occurred in the medial head and body and along the lateral side
of the hippocampi. The highest significance was found in the
medial head of the right hippocampus in the AD group (Fig. 3A
and online 3D mesh, see Methods). Figures 2 and 3 suggest a
similar pattern of atrophy rate among the three groups with
different magnitudes that were further examined.
Region of Interest Analysis

To further test for differences between pMCI and sMCI, we
performed a Region of Interest (ROI) analysis of the atrophy rate.
The ROIs were used to increase the sensitivity to distinguish
patients with stable MCI from progressive ones. We defined the
ROIs as clusters of vertices with significantly higher atrophy rates
in AD than in control subjects; which make them independently
defined from the MCI dataset. A total of four ROIs were defined
both in the right (R1R-R4R) and left (R1L-R4L) hippocampi. R1
is the cluster of vertices in the medial side of the hippocampal
head, R2 corresponds to the cluster in the medial side of the body,
R3 represents the cluster in the lateral side of head, whereas R4
contains the clusters along the lateral side of the hippocampal
body and tail. The ROIs therefore do not correspond to a specific
hippocampal subfield derived from the high-resolution atlas (see
section Hippocampal subfields).
Within each region, we computed the average atrophy rate for
each MCI patient. The stable and progressive MCI groups were
compared statistically using Wilcoxon test (significance level was
adjusted for the number of ROIs with Bonferroni correction).
Receiver Operating Characteristic (ROC) analysis was then
performed using the R package pROC version 1.4.4 [28].
Group Differences
The significantly higher rate of atrophy in patients with AD
than in controls (Fig. 4A and online 3D mesh, see Methods) was
found mainly in the medial part of the head and body and along
the lateral side of the hippocampi (Wilcoxon test, pv0.05 corr.). A
higher atrophy rate in the MCI group as compared to controls
(Fig. 4B) was found in much smaller regions, mainly on the lateral
side of the hippocampal head. Finally, significant differences
between the AD and MCI groups (Fig. 4C) were found in a few
vertices along the lateral side.
Results
Stable versus Progressive Mild Cognitive Impairment
Hippocampal Volume Change
120
The difference in atrophy rate between the pMCI and sMCI

groups (Fig. 5) was only assessed within the regions showing
significantly higher atrophy rate in AD compared with control in
order to increase the sensitivity. These comparisons revealed
significantly higher rates in pMCI in all but one ROI (Wilcoxon
test, R1R: W = 518, p-value = 0.001951; R1L: W = 497, pvalue = 0.001004; R2R: W = 563, p-value = 0.007208; R2L:
We first examined the difference in hippocampal volume

between MRI_acq1 and MRI_acq2. Estimations of hippocampal
volumes (in mm3) for MRI_acq1 and the volume change between
MRI_acq1 and MRI_acq2 (in mm3/year) are listed in Table 2.
We found significant hippocampal volume loss rates in each group
(Table 2). However, the loss was greater in patients with AD than

Figure 1. Illustration of the methods. (A) On the left side of this panel, we illustrate the triangulated surface of the template hippocampus and
the manually defined seven points (in red). On the right side, the seven points drawn on the subjects hippocampus are shown in blue, and the ICP
algorithm is illustrated which minimises the sum of distances (di) between the red and blue points. (B) Illustration of the deformation field mapping
onto the triangulated hippocampal surface resulting in different colours for inward (blue) and outward (orange) deformation.
W = 570, p-value = 0.00871; R3R: W = 453, p-value = 0.0002236;

R3L: W = 519, p-value = 0.002011; R4R: W = 737, n.s; R4L:
W = 543, p-value = 0.004110). The largest difference was found in
the R3R ROI, which represents the antero-lateral part of the right
hippocampus. Within this ROI, we performed an ROC analysis to
determine how well the atrophy rate of this region is able to
distinguish between pMCI and sMCI. The ROC analysis revealed
72.4% accuracy in discriminating the two groups, with 70.5%

specificity and 74.4% sensitivity (area under the curve: 78.1%).
Discussion
In the present study, we used two MRI scans to map the rate of
hippocampal atrophy in patients with AD and MCI and in healthy
elderly controls. We also measured the total hippocampal volume
121
Figure 2. Mean atrophy rate. Mean atrophy rate in mm/year for the left and right hippocampi in patients with AD (A), MCI (B) and controls (C). On
the right hippocampus, projected borders of subfields CA1, CA23, dentate gyrus (DG) and subiculum (Sub) are indicated with grey outlines.
122

Table 2. Summary of hippocampal volumetry.
Group
Volume MRI_acq1 (R/L)
Volume change per year (R/L)
Statistics and p-value (R/L)
2*Control (n = 85)
2402 (476)/
29 (30)/
V = 774, p-value = 1.975e-06/
2059 (404)
26 (22)
V = 992, p-value = 0.0001267
2181 (468)/
220 (31)/
V = 815, p-value = 2.079e-09/
1946 (431)
210 (30)
V = 1360, p-value = 3.115e-05
2129 (446)/
228 (41)/
V = 479, p-value = 1.403e-10/
1805 (397)
214 (34)
V = 978, p-value = 8.489e-06
2*MCI (n = 102)
2*AD (n = 90)
Average hippocampal volume in mm3 (SD) at the first scan, average volume loss in mm3/year between the two scans for right and left (R/L) hippocampi (SD), and the
statistical tests on volume loss (Wilcoxon test) in each group.
and compared its decrease among the three groups. We found the
lowest baseline volume in the AD group, and the greatest in the
group of healthy elderlies. The average hippocampal volume of
patients with MCI was in between the other two groups. Looking
at the volume loss over time, we found the fastest loss in the AD
group. This was significantly higher than the loss in the normal
group both in the left and the right hippocampi. The difference in
volume loss between the AD and MCI groups was significant only
in the right side. Similarly, only the right hippocampus showed
greater volume loss in the MCI group than in controls.
Comparison of the atrophy rates among the three subject groups
revealed similar patterns. The regions with the highest rate of
atrophy were located on the medial side of the hippocampal head
and body and along the lateral side. These regions also showed
significantly higher atrophy rates in patients with AD compared
with controls. The difference in hippocampal atrophy rate
between MCI and controls was located at the lateral side of the
right hippocampal head, whereas the higher atrophy rate in AD
compared to MCI was mainly found on the lateral side of the right
hippocampal body and tail. To relate these regions to the
histological subparts of the hippocampus, we used the atlas of the
human hippocampus segmented from high-resolution MRIs [27].
Based on this atlas, the fastest atrophy occurred in the CA1 zone.
Similarly, the strongest differences between the groups were
localised in the CA1 zone and in the subiculum.
Our volumetric measurements are in accordance with previous
studies measuring the hippocampal volume in different populations, namely sequential decrease in volume among healthy
elderlies, MCI and AD patients [29]. We also found smaller
volume in the left side compared to the right, as did many groups
previously [3032]. Our results are also in line with studies
examining the volume loss over time [11,31,3337]. They
reported the greatest annual percent change in the hippocampal
volume in patients with AD. This was significantly higher than
that in patients with MCI and in normal subjects. Leung et al. [37]
showed that the volume loss was even accelerating in the MCI
group over time.
Although the difference in hippocampal volume had the same
trend in many studies previously, the volume itself differed
substantially. Groups, using the ADNI database of patients with
AD and MCI and healthy elderly controls, reported hippocampal
volumes in AD varying between 1600 [35] and 3000 mm3 [20]. In
the present study, the hippocampal volume of AD patients was
about 2000 mm3. The remarkable difference among studies might
derive from the segmentation of the template hippocampus used to
extract the structure in each individual. There are different
recommendations for hippocampal segmentation which include
different amount of the subiculum and the tail of the hippocampus
[38]. It can also differ how far anterior the hippocampus is

segmented inferior to the amygdala. Here we followed the
recommendations of Franko et al. [23] and Insausti and Amaral
[24]. These studies described landmarks visible on MRI which are
based on the histological examination and post-mortem MRI of
the same brains.
Recently, several studies have focused on 3D shape analyses and
detailed hippocampal atrophy mapping, and compared these
maps among healthy controls and patients with MCI and AD
[20,21,32,3943]. Most of them used a single MRI scan and
looked for differences in the hippocampal shape between healthy
elderly and patients [20,39,40,4447], referring to this shape
difference as atrophy. However, this requires the assumption that
the patient group had the same hippocampal shape as the control
group prior to the disease process, which might not be the case if
the number of subjects is not sufficiently high. Importantly, more
and more studies measure the atrophy using a longitudinal MRI
dataset, allowing the visualisation of the hippocampal atrophy
within each single subject over time [15,19,21,32,42]. This
method does not require a template hippocampus computed by
averaging of healthy subjects brains, therefore it is more robust for
individual variability in the hippocampal shape. However, the
methods could be biased by asymmetric deformation mapping
[26,48], which could explain the contradicting findings in the
previous studies. We therefore examined the hippocampal atrophy
performing an unbiased symmetric deformation field estimation.
One of the first studies comparing the hippocampal atrophy rate
in patients with AD to that in healthy subjects was performed by
Thompson et al. [19]. They reported similar atrophy rate patterns
between the two groups but found significantly higher atrophy
rates on the lateral side of the left hippocampal head in AD
patients. We also found similar atrophy rate patterns among the
groups; however, we found significantly higher rates in AD
patients compared to controls not only on the lateral side of the
head but also on the medial part of the head and body and the
lateral side of the hippocampal body and tail. The hippocampal
atrophy rate maps reported by other groups [21,32,42] differ more
from our results. Morra et al. [21] mainly found atrophy on the
superior and inferior parts of the body and on the tail of the
hippocampus of AD patients, whereas in our study, the atrophy
rate was higher on the lateral and medial sides relatively sparing
the superior and inferior surfaces. However, when comparing the
patients to healthy controls, the significantly different atrophy was
located mainly on the inferior part of the head and on the lateral
and medial sides of the hippocampus, which is closer to our
findings. However, another study from the same group [20]
reported significant difference between AD patients and controls
on almost the entire surface of the hippocampus sparing only a line 123
6
Figure 3. Significant atrophy rates. Statistical maps showing the -log(p-values) for significant (pv0.05 corr.) atrophy rates for the left and right
hippocampi in patients with AD (A), MCI (B) and controls (C). Significance level (pv0.05 corr.) is indicated in the colour bar.
124

Figure 4. Statistical maps of atrophy rate differences among groups. Statistical maps showing significantly higher atrophy rate in AD versus
control (A), MCI versus control (B), and AD versus MCI (C). Significance level (pv0.05 corr.) is indicated in the colour bar.
125
Figure 5. ROI analysis of atrophy rates in progressive and stable MCI. Bar plots of atrophy rates in the left and right hippocampal ROIs
(shown in Figure 4) for pMCI (n = 39) and sMCI (n = 44). *: pv0.05, **: pv0.001 Bonferroni adjusted p-values.
126
along the middle of the superior and inferior surfaces. Wang et al.
[32,39] compared patients with MCI (CDR = 0.5) to healthy
controls. They found more inward deformation in patients on the
superior side of the head and the inferior and lateral parts of the
hippocampal body and tail. They also reported the inward
deformation of these regions to predict conversion from healthy
state to MCI [49]. Apostolova et al. [42] compared subjects who
remained cognitively normal with those who converted to MCI or
AD and showed significant differences in hippocampal atrophy
rate in all parts of the hippocampus without subfield specificity
(CA13 subfields and subiculum).
Although Gerardin et al. [40] used a single scan to examine the
hippocampal shape, the topography of atrophy found in the
present study is the most like their results. They found
morphological differences between the hippocampi of patients
with AD and those of controls, predominantly on the medial part
of the head and the lateral part of the body and tail. However, the
use of a single scan prevented the investigation of atrophy
evolution. Our data suggest that the morphological differences
among groups in their study may be the consequence of different
atrophy rates. Costafreda et al. [43] also used a single scan to
predict conversion from MCI to AD using a baseline atrophic
phenotype. Similarly to our findings, the atrophy of the anterior
part of the CA1 field best predicted the conversion, with 80%
accuracy. We also found the lateral side of the hippocampal head
(corresponding to CA1) to be the best predictor of the progression
to AD from MCI based on the atrophy rate. Although the
accuracy is slightly lower, our method can be used to identify
patients at high risk of progression. This can help identify patients
most likely to be helped by treatments that are mainly effective in
the early phase of AD.
We found similar patterns of atrophy rate among the three
subject groups which might suggest that certain hippocampal
subregions are more prone to atrophy caused by normal or
pathological ageing. However, the current method cannot
distinguish between different sources of atrophy (underlying

mechanisms of cell death). Therefore, we cannot speculate
whether the similarity in atrophy pattern in AD compared to
the controls would indicate that the disease simply accelerates the
normal ageing. Braak and Braak [5] found that tau (neurofibrillary
tangle and neuropil thread) deposition in the hippocampus first
occurs on the medial and lateral side of the CA1 field in the early
phase of AD (stage II, their Fig. 4); these regions are also more
affected in the later phases than the middle portion of the CA
(superior portion of the 3D hippocampus, see Methods) or the
subiculum (infero-medial portion of the 3D hippocampus).
Importantly, these regions that have the highest burden of
neurofibrillary tangle and neuropil thread deposition correspond
to those showing the highest atrophy rate in our study. This good
correspondance between our results and the well-established
neuropathology in AD supports a biological interpretation of the
current mapping of atrophy rate.
Longitudinal studies are also necessary to evaluate atrophy
progression at the individual level, which is the best way to
objectively measure treatment effects. Jack and colleagues
previously demonstrated that higher annual hippocampal volume
loss correlated with worsening of the clinical status [11]. However,
when comparing treated and untreated AD patients, Fox et al.
[34] reported non-significant hippocampal volume change.
Similarly, Wang et al. [41] did not find significant differences in
hippocampal volume or surface deformation between treated and
untreated MCI patients. Their method of averaging the deformation within each subfield assumes homogeneous atrophy. This
might decreases the sensitivity to mild treatment effects. The
present method is more sensitive to local deformation, which
makes it more appropriate for use in clinical trials.
Beside its high sensitivity to local deformation, our method is
also free of biases arising from asymmetric global normalization
[26]. In a very recent study, Lorenzi and colleagues [48]
developed another local symmetric registration algorithm robust

to intensity bias. Their method will be useful for future

longitudinal studies. Even though we used a semi-automated
segmentation of the hippocampus at baseline, this step could be
replaced with a fully-automated method. Several groups have
developed fully-automated hippocampal segmentation tools [50
52] and some are recently available using FIRST/FSL and
FreeSurfer softwares. Although our method is more timeconsuming for large cohorts of patients and requires anatomical
knowledge, it enabled us to use hippocampal boundaries as
described by Insausti and Amaral [24].
In conclusion, our MRI-based regional hippocampal atrophy
maps are in high agreement with the known AD histopathology.
This shows that longitudinal MRI and particularly the measurement of local hippocampal atrophy rate are reliable methods to
investigate AD. Additionally, we describe an unbiased method to
objectively evaluate AD progression that can be used in clinical
trials to test novel disease-modifying drugs and measure their
efficacy.
Acknowledgments
The authors are indebted to T.E Cope for comments on earlier versions of
the manuscript.
investigators can be found at: http://adni.loni.ucla.edu/wp-content/
uploads/how_to_ apply/ADNI_Acknowledgement_List.pdf.
Conceived and designed the experiments: EF OJ. Performed the
experiments: EF OJ. Analyzed the data: EF OJ. Contributed reagents/
materials/analysis tools: EF OJ. Wrote the paper: EF OJ.
References
Automated 3d mapping of hippocampal atrophy and its clinical correlates in 400
subjects with alzheimers disease, mild cognitive impairment, and elderly
controls. Hum Brain Mapp 30: 276688.
Automated mapping of hippocampal atrophy in 1-year repeat mri data from 490
subjects with alzheimers disease, mild cognitive impairment, and elderly
controls. Neuroimage 45: S315.
22. Franko E, Joly O (2010) Mapping changes in the hippocampal shape over time
in alzheimers disease. Program No 75620 2010 Neuroscience Meeting Planner
San Diego, CA: Society for Neuroscience, 2010 Online.
23. Franko E, Insausti AM, Artacho-Perula E, Insausti R, Chavoix C (2012)
Identification of the human medial temporal lobe regions on magnetic resonance
images. Hum Brain Mapp : In Press, DOI: 10.1002/hbm.22170.
24. Insausti R, Amaral D (2004) Hippocampal formation. The human nervous
system: 871914.
25. Pluta J, Avants BB, Glynn S, Awate S, Gee JC, et al. (2009) Appearance and
incomplete label matching for diffeomorphic template based hippocampus
segmentation. Hippocampus 19: 56571.
26. Yushkevich PA, Avants BB, Das SR, Pluta J, Altinay M, et al. (2010) Bias in
estimation of hippocampal atrophy using deformation-based morphometry
arises from asymmetric global normalization: an illustration in adni 3 t mri data.
27. Yushkevich PA, Avants BB, Pluta J, Das S, Minkoff D, et al. (2009) A highresolution computational atlas of the human hippocampus from postmortem
magnetic resonance imaging at 9.4 t. Neuroimage 44: 38598.
28. Robin X, Turck N, Hainard A, Tiberti N, Lisacek F, et al. (2011) proc: an opensource package for r and s+ to analyze and compare roc curves. BMC
Bioinformatics 12: 77.
29. Convit A, Leon MJD, Tarshish C, Santi SD, Tsui W, et al. (1997) Specific
hippocampal volume reductions in individuals at risk for alzheimers disease.
30. Jack CR, Twomey CK, Zinsmeister AR, Sharbrough FW, Petersen RC, et al.
(1989) Anterior temporal lobes and hippocampal formations: normative
volumetric measurements from mr images in young adults. Radiology 172:
549554.
31. Laakso MP, Lehtovirta M, Partanen K, Riekkinen PJ, Soininen H (2000)
Hippocampus in alzheimers disease: a 3-year follow-up mri study. Biol
32. Wang L, Swank JS, Glick IE, Gado MH, Miller MI, et al. (2003) Changes in
hippocampal volume and shape across time distinguish dementia of the
alzheimer type from healthy aging. Neuroimage 20: 66782.
33. Jack CR, Shiung MM, Gunter JL, OBrien PC, Weigand SD, et al. (2004)
Comparison of different mri brain atrophy rate measures with clinical disease
progression in ad. Neurology 62: 591600.
34. Fox NC, Black RS, Gilman S, Rossor MN, Griffth SG, et al. (2005) Effects of
abeta immunization (an1792) on mri measures of cerebral volume in alzheimer
disease. Neurology 64: 15631572.
35. Schuff N,Woerner N, Boreta L, Kornfield T, Shaw LM, et al. (2009) Mri of
hippocampal volume loss in early alzheimers disease in relation to apoe
genotype and biomarkers. Brain 132: 10671077.
36. Macdonald KE, Bartlett JW, Leung KK, Ourselin S, Barnes J, et al. (2012) The
value of hippocampal and temporal horn volumes and rates of change in
predicting future conversion to ad. Alzheimer Dis Assoc Disord.
37. Leung KK, Bartlett JW, Barnes J, Manning EN, Ourselin S, et al. (2013)
Cerebral atrophy in mild cognitive impairment and alzheimer disease: Rates and
acceleration. Neurology 80: 648654.
1. Ferri CP, Prince M, Brayne C, Brodaty H, Fratiglioni L, et al. (2005) Global

prevalence of dementia: a delphi consensus study. Lancet 366: 21127.
2. Cummings JL (2004) Treatment of alzheimers disease: current and future
therapeutic approaches. Rev Neurol Dis 1: 609.
4. Braak H, Braak E, Bohl J (1993) Staging of alzheimer-related cortical
destruction. Eur Neurol 33: 4038.
5. Braak H, Braak E (1995) Staging of alzheimers disease-related neurofibrillary
changes. Neurobiol Aging 16: 2718; discussion 27884.
6. Price JL, Ko AI, Wade MJ, Tsou SK, McKeel DW, et al. (2001) Neuron number
in the entorhinal cortex and ca1 in preclinical alzheimer disease. Arch Neurol
58: 1395402.
7. Bobinski M, Wegiel J, Wisniewski HM, Tarnawski M, Bobinski M, et al. (1996)
Neurofibrillary pathologycorrelation with hippocampal formation atrophy in
alzheimer disease. Neurobiol Aging 17: 90919.
8. Juottonen K, Laakso MP, Insausti R, Lehtovirta M, Pitkanen A, et al. (1998)
Volumes of the entorhinal and perirhinal cortices in alzheimers disease.
9. Convit A, de Asis J, de Leon MJ, Tarshish CY, De Santi S, et al. (2000) Atrophy
of the medial occipitotemporal, inferior, and middle temporal gyri in nondemented elderly predict decline to alzheimers disease. Neurobiol Aging 21: 19
26.
10. Chetelat G, Baron JC (2003) Early diagnosis of alzheimers disease: contribution
of structural neuroimaging. Neuroimage 18: 52541.
11. Jack CR, Petersen RC, Xu Y, OBrien PC, Smith GE, et al. (2000) Rates of
hippocampal atrophy correlate with change in clinical status in aging and ad.
12. Fox NC, Cousens S, Scahill R, Harvey RJ, Rossor MN (2000) Using serial
registered brain magnetic resonance imaging to measure disease progression in
alzheimer disease: power calculations and estimates of sample size to detect
treatment effects. Arch Neurol 57: 33944.
13. van de Pol LA, Barnes J, Scahill RI, Frost C, Lewis EB, et al. (2007) Improved
reliability of hippocampal atrophy rate measurement in mild cognitive
impairment using uid registration. Neuroimage 34: 10361041.
14. Misra C, Fan Y, Davatzikos C (2009) Baseline and longitudinal patterns of brain
atrophy in mci patients, and their use in prediction of short-term conversion to
ad: results from adni. Neuroimage 44: 14151422.
15. Risacher SL, Shen L, West JD, Kim S, McDonald BC, et al. (2010) Longitudinal
mri atrophy biomarkers: relationship to conversion in the adni cohort. Neurobiol
Aging 31: 14011418.
Measurement of hippocampal atrophy using 4d graph-cut segmentation:
application to adni. Neuroimage 52: 109118.
17. Leung KK, Barnes J, Ridgway GR, Bartlett JW, Clarkson MJ, et al. (2010)
Automated crosssectional and longitudinal hippocampal volume measurement
in mild cognitive impairment and alzheimers disease. Neuroimage 51: 1345
1359.
18. Henneman WJP, Sluimer JD, Barnes J, van der Flier WM, Sluimer IC, et al.
(2009) Hippocampal atrophy rates in alzheimer disease: added value over whole
brain volume measures. Neurology 72: 9991007.
19. Thompson PM, Hayashi KM, De Zubicaray GI, Janke AL, Rose SE, et al.
(2004) Mapping hippocampal and ventricular change in alzheimer disease.
10
127
38. Konrad C, Ukas T, Nebel C, Arolt V, Toga AW, et al. (2009) Defining the
human hippocampus in cerebral magnetic resonance imagesan overview of
current segmentation protocols. Neuroimage 47: 118595.
39. Wang L, Miller JP, Gado MH, McKeel DW, Rothermich M, et al. (2006)
Abnormalities of hippocampal surface structure in very mild dementia of the
alzheimer type. Neuroimage 30: 5260.
alzheimers disease and mild cognitive impairment from normal aging.
41. Wang L, Harms MP, Staggs JM, Xiong C, Morris JC, et al. (2010) Donepezil
treatment and changes in hippocampal structure in very mild alzheimer disease.
42. Apostolova LG, Mosconi L, Thompson PM, Green AE, Hwang KS, et al. (2010)
Subregional hippocampal atrophy predicts alzheimers dementia in the
cognitively normal. Neurobiol Aging 31: 107788.
43. Costafreda SG, Dinov ID, Tu Z, Shi Y, Liu CY, et al. (2011) Automated
hippocampal shape analysis predicts the onset of dementia in mild cognitive
impairment. Neuroimage 56: 2129.
44. Gutman B, Wang Y, Morra J, Toga AW, Thompson PM (2009) Disease
classification with hippocampal shape invariants. Hippocampus 19: 572578.
45. Frisoni GB, Sabattoli F, Lee AD, Dutton RA, Toga AW, et al. (2006) In vivo
neuropathology of the hippocampal formation in ad: a radial mapping mr-based
study. Neuroimage 32: 104110.
46. Frisoni GB, Ganzola R, Canu E, Rb U, Pizzini FB, et al. (2008) Mapping local
hippocampal changes in alzheimers disease and normal ageing with mri at 3
tesla. Brain 131: 32663276.
47. Pluta J, Yushkevich P, Das S, Wolk D (2012) In vivo analysis of hippocampal
subfield atrophy in mild cognitive impairment via semi-automatic segmentation
of t2-weighted mri. J Alzheimers Dis 31: 8599.
48. Lorenzi M, Ayache N, Frisoni GB, Pennec X, for the Alzheimers Disease
Neuroimaging Initiative (ADNI) (2013) Lcc-demons: A robust and accurate
diffeomorphic registration algorithm. Neuroimage.
49. Csernansky JG, Wang L, Swank J, Miller JP, Gado M, et al. (2005) Preclinical
detection of alzheimers disease: hippocampal shape and volume predict
dementia onset in the elderly. Neuroimage 25: 783792.
50. Leemput KV, Bakkour A, Benner T, Wiggins G, Wald LL, et al. (2009)
Automated segmentation of hippocampal subfields from ultra-high resolution
in vivo mri. Hippocampus 19: 549557.
51. Chupin M, Mukuna-Bantumbakulu AR, Hasboun D, Bardinet E, Baillet S, et al.
(2007) Anatomically constrained region deformation for the automated
segmentation of the hippocampus and the amygdala: Method and validation
on controls and patients with alzheimers disease. Neuroimage 34: 9961019.
52. Chupin M, Grardin E, Cuingnet R, Boutet C, Lemieux L, et al. (2009) Fully
automatic hippocampus segmentation and classification in alzheimers disease
and mild cognitive impairment applied on data from adni. Hippocampus 19:
579587.
128
11
Other titles by iMedPub:

Social Medicine in the 21st Century by Samuel Barrack.
World Health Report 2012: No Health Without Research by Samuel
Barrack.
Quality design in Anatomical Pathology by Anil Malleshi Betigeri.
Escherichia coli infections by Viroj Wiwanitkit.
Atlas of Biomarkers for Alzheimers disease by Manuel Menendez.
130
Current diagnosis of Alzheimers diseaserelies largely on diagnostic criteria

based on documenting clinical features. In the past few years, the AD
research enterprise has developed improved techniques and technologies
aimed at early diagnosis and detection of AD.
The field has seen remarkable growth in research using clinical assessment,
cerebrospinal fluid-based biomarkers, magnetic resonance imaging
measurements of brain volume, functional imaging and molecular imaging
to diagnose AD at its earliest stages.Clinicians are bombarded with vast
amounts of information from different sources and it is good to compile and
arrange the most outstanding science in the field.

Neuroimaging Biomarkers in Alzheimer's Disease

Uploaded by

Copyright:

Neuroimaging Biomarkers in Alzheimer's Disease

Uploaded by

Document Information

Original Title

Copyright

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Neuroimaging Biomarkers in Alzheimer's Disease

Uploaded by

Copyright:

Thematic Collections

Title: Neuroimaging biomarkers

Cover design and Layout: [email protected]

First edition: 2013

Neuroimaging biomarkers in Alzheimers disease

Preface for Biomarkers for

Diagnosed Mild Cognitive Impairment

Multi-Method Analysis of MRI Images

Combined Evaluation of FDG-PET

Plasma Based Markers of [11C]

Generative FDG-PET and MRI Model

Diffusion Tensor Metrics as Biomarkers in Alzheimers

measures, can remain relatively unperturbed. Conflicting results,

There is presently considerable interest in trying to expedite

PLOS ONE | www.plosone.org

November 2012 | Volume 7 | Issue 11 | e49072

Neuroimaging biomarkers in Alzheimers disease

Very mild Alzheimers

Age at imaging, years

Attention & Orientation/18

disease i.e. the corpus callosum [3,4,7,8,11,12,13,15,16]. The aim

Table 2. Cognitive profile evolution of a group of early-stage

The study was approved by the Cambridgeshire 2 Research

Global Cognition MMSE/30

ACE-R Subscores Attention &

Visuospatial/16 14.7 (1.5)

Cognitive measures are given as mean (SD).

recruited and were screened to exclude neurological or major

Figure 1. Cognitive status. Depiction of average cognitive profiles

PLOS ONE | www.plosone.org

November 2012 | Volume 7 | Issue 11 | e49072

Diffusion Tensor Biomarkers in AD

have probable Alzheimers disease by confirming progressive

All patients were scanned within an average of 1.3 months

Data Processing and Analysis

Diffusion tensor parametric maps. The Oxford Centre

PLOS ONE | www.plosone.org

November 2012 | Volume 7 | Issue 11 | e49072

Neuroimaging biomarkers in Alzheimers disease

PLOS ONE | www.plosone.org

(FNIRT v1.0). Allpatients (N = 43) and controls (N = 26)warped

November 2012 | Volume 7 | Issue 11 | e49072

Diffusion Tensor Biomarkers in AD

skeletonisation vectors. All statistical maps were thresholded at

An alternate hypothesis to explain this phenomenon, however, was

November 2012 | Volume 7 | Issue 11 | e49072

Neuroimaging biomarkers in Alzheimers disease

Each T1-weighted structural volume was affine-registered to its

Figure 5. Cross-sectional results in the mid-sagittal corpus

PLOS ONE | www.plosone.org

November 2012 | Volume 7 | Issue 11 | e49072

Diffusion Tensor Biomarkers in AD

Alzheimers disease (N = 43) vs. Control (N = 26)

Results are given as Wilcoxon rank-sum Z-statistic.

samples (i.e. patients versus controls or very mild versus mild

to controls. This experiment was treated as a family (n = 28 tests);

Mild Alzheimers (N = 22) vs.

Mild Alzheimers (N = 22) vs. Very

PLOS ONE | www.plosone.org

November 2012 | Volume 7 | Issue 11 | e49072

Neuroimaging biomarkers in Alzheimers disease

Mid-sagittal Corpus Callosum

across the right hemisphere. FA changes were similarly distributed,

Title: Neuroimaging biomarkers

Cover design and Layout: [email protected]