1-s2.0-S1053811924000648-main

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

NeuroImage 290 (2024) 120569

Contents lists available at ScienceDirect

NeuroImage
journal homepage: www.elsevier.com/locate/ynimg

Spatial correspondence of cortical activity measured with whole head


fNIRS and fMRI: Toward clinical use within subject
Anthony Zinos a, Julie C. Wagner a, Scott A. Beardsley a, b, *, Wei-Liang Chen c, Lisa Conant b,
Marsha Malloy b, d, Joseph Heffernan b, Brendan Quirk b, Robert Prost b, Mohit Maheshwari e,
Jeffrey Sugar b, Harry T. Whelan b, d
a
Department of Biomedical Engineering, Marquette University and Medical College of Wisconsin, Milwaukee, WI, USA
b
Department of Neurology, Medical College of Wisconsin, Milwaukee, WI, USA
c
Center for Neuroscience Research, Children’s National Medical Center, George Washington University, Washington DC, USA
d
Department of Neurology, Children’s Wisconsin, Milwaukee, WI, USA
e
Department of Radiology, Children’s Wisconsin, Milwaukee, WI, USA

A R T I C L E I N F O A B S T R A C T

Keywords: Functional near infrared spectroscopy (fNIRS) and functional magnetic resonance imaging (fMRI) both measure
fNIRS the hemodynamic response, and so both imaging modalities are expected to have a strong correspondence in
fMRI regions of cortex adjacent to the scalp. To assess whether fNIRS can be used clinically in a manner similar to
DOT
fMRI, 22 healthy adult participants underwent same-day fMRI and whole-head fNIRS testing while they per­
Motor cortex
Visual cortex
formed separate motor (finger tapping) and visual (flashing checkerboard) tasks. Analyses were conducted
Functional cortical activity within and across subjects for each imaging approach, and regions of significant task-related activity were
compared on the cortical surface. The spatial correspondence between fNIRS and fMRI detection of task-related
activity was good in terms of true positive rate, with fNIRS overlap of up to 68 % of the fMRI for analyses across
subjects (group analysis) and an average overlap of up to 47.25 % for individual analyses within subject. At the
group level, the positive predictive value of fNIRS was 51 % relative to fMRI. The positive predictive value for
within subject analyses was lower (41.5 %), reflecting the presence of significant fNIRS activity in regions
without significant fMRI activity. This could reflect task-correlated sources of physiologic noise and/or differ­
ences in the sensitivity of fNIRS and fMRI measures to changes in separate (vs. combined) measures of oxy and
de-oxyhemoglobin. The results suggest whole-head fNIRS as a noninvasive imaging modality with promising
clinical utility for the functional assessment of brain activity in superficial regions of cortex physically adjacent to
the skull.

1. Introduction NIRS is a portable, noninvasive, and cost-effective imaging modality,


that can be used at bedside to continuously monitor brain activity in real
Functional near infrared spectroscopy (fNIRS) and functional mag­ time. Despite being first described by Jöbsis over 40 years ago (Jobsis,
netic resonance imaging (fMRI) both measure the hemodynamic 1977), and its adoption as standard of care for hemodynamic monitoring
response, which is a localized change in cerebral blood flow in response of the brain and other vital organs in surgery and critical care settings,
to neuronal activity. The hemodynamic response arises in reaction to the multichannel fNIRS has not yet seen widespread adoption in clinical
increased metabolic demands of neuronal activity within brain regions, neuroscience (Obrig, 2014; Hong and Yaqub, 2019). While usage of
i.e., neurovascular coupling (Fox and Raichle, 1984). While fMRI mea­ fNIRS and diffuse optical tomography (DOT) as research tools to char­
sures the hemodynamic response via net differences in the paramagnetic acterize brain activity across subjects have steadily increased, clinical
properties of oxy- versus deoxyhemoglobin, fNIRS uses changes in applications have been limited, due in part to the relatively lower signal
near-infrared light absorbance to separately measure changes in cere­ to noise ratio and uncertainty in the degree of spatial specificity for
bral oxy- and deoxyhemoglobin via the Beer-Lambert Law. analyses performed within subject (Chen et al., 2020). To the extent that

* Corresponding author.
E-mail address: [email protected] (S.A. Beardsley).

https://doi.org/10.1016/j.neuroimage.2024.120569
Received 29 August 2023; Received in revised form 15 December 2023; Accepted 7 March 2024
Available online 8 March 2024
1053-8119/© 2024 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
A. Zinos et al. NeuroImage 290 (2024) 120569

fNIRS detection of brain activity in superficial regions of cortex may be sat upright in a darkened room while viewing a 44 cm computer display
comparable to fMRI, it could be a promising alternative in cases where a placed at eye level at approximately 60 cm away.
patient’s medical condition or history make fMRI procedures unsuitable Functional and structural MR images were obtained using a GE
or unsafe to complete. Healthcare 3T Signa Premier MR scanner. High-resolution whole brain
Some recent fNIRS studies have focused on using group analysis to anatomical images were obtained at the beginning of the imaging ses­
validate fNIRS with fMRI at the channel (Donizete de Faria et al., 2020; sion using T1-weighted gradient recall images (176 slices, TR = 8 ms, TE
Klein et al., 2022) and source levels (Hernández-Martin et al., 2017; = 3.1 ms, flip angle = 12◦ , 1 × 1 × 1 mm3 voxels). Functional MR images
Huppert et al., 2017; Wijeakumar et al., 2017). Hernández-Martin and consisted of axially oriented T2-weighted echo planar images covering
colleagues looked at a source level comparison during a set of cognitive the whole brain (41 slices, TR = 2.0 s, TE = 23 ms, flip angle = 77◦ , 3.5 ×
(arithmetic) tasks and found voxel-wise overlap between sources of 3.5 × 3.5 mm3 voxels). The number of functional volumes varied for
fNIRS and fMRI activity that varied with task from 5 % to 59 % each task: 205 for the Motor task and 240 for the Visual task. During the
(Hernández-Martin et al., 2017). In another study, source level com­ fMRI session, participants wore the same cap used during fNIRS data
parisons during parametric median nerve stimulation showed a mod­ acquisition with vitamin E capsules placed in select channel locations to
erate spatial correlation (R = 0.57) between oxyhemoglobin fNIRS verify channel registration on the participant’s head.
measurements and the blood oxygen level dependent (BOLD) fMRI
response (Huppert et al., 2017), with a 9 mm minimum displacement 2.3. Experimental tasks
between the center of mass activity in superficial regions of cortex. In
these and other validation studies, fNIRS optodes were concentrated on Participants took part in same day fNIRS and fMRI experimental
a specific region of interest, limiting spatial comparisons of functional sessions in which they completed separate localizer tasks assessing
activity only to regions where activity was expected, and results were motor and visual function. Each task was presented in a block paradigm
analyzed at the group level (i.e., across subjects). To be clinically useful, consisting of task periods (20 s each) interleaved with rest periods (20 s
fNIRS needs to show comparable detection of functional activity across each). Motor cortices were functionally localized using a sequential
the cortical surface, and at an individual level. finger tapping task (i.e., Motor task) in which participants tapped their
This study aimed to compare source-localized, task-related fNIRS to thumb to each finger starting with their index finger going towards their
same-day fMRI. Temporal correspondence has been well studied in fifth digit and then in reverse as quickly as possible while still hitting
previous work (Cui et al., 2011; Noah et al., 2015; Strangman et al., each individual digit. Participants tapped their fingers sequentially
2002) which has consistently shown significant temporal correlations while the word for the hand cue was displayed (20-second). When the
between fNIRS and fMRI, though the strength of these correlations ap­ word RIGHT appeared on the screen, participants tapped the fingers on
pears to vary with scalp-brain distance (Cui et al., 2011). This study their right hand, and when the word LEFT was displayed, they tapped
instead centered on the spatial correspondence between fNIRS and fMRI, the fingers on their left hand. Following each tapping period, partici­
with a focus on clinical applications that require spatially accurate pants were presented with a 20 second rest period and asked to fixate a
localization of brain activity. In contrast to previous studies that used an crosshair located at the center of the screen while performing no
ROI based montage, this study used a whole-head montage that covered movement. All participants completed 10 blocks of the Motor task for
even cortical regions distant from the expected task response. This each hand.
layout was intended to provide better estimates of the false positive rate The visual cortex of each subject was functionally localized using a
of fNIRS in clinical contexts where prior information of source locations flashing checkerboard (i.e., Visual task). During the task, participants
might be unavailable. Comparisons were performed across the cortical were instructed to fixate on a yellow dot at the center of the screen while
surface at a group and individual level to help assess the clinical po­ a flickering checkered wedge pattern of high contrast black and white
tential of fNIRS for identifying functionally active regions of cortex. wedges was displayed (DeYoe et al., 2015). The checkerboard pattern
occupied 40◦ of the participants’ field of view. The field of view was
2. Materials and methods controlled to be identical in both the fNIRS and fMRI experiments.
Seventeen concentric circles were divided into 20 wedges of 18◦ each
2.1. Participants that counterphase flickered at 8 Hz. The pattern appeared for 20 s,
followed by a 20 second rest period during which only the yellow dot
Following a screening process where subjects performed a simple was visible against a medium grey background. A subset of 9 partici­
motor task to determine if a strong task-related hemodynamic response pants completed 12 cycles of the Visual task.
was detected with fNIRS, 22 healthy adults (39.1 ± 15.1 years, 6 male) Experimental presentation of the task sequences and stimuli were
participated in a same-day fNIRS-fMRI study. All participants were performed using PsychoPy3 (Peirce et al., 2019). The order of experi­
right-hand dominant as determined by the Edinburgh Handedness In­ ments was randomized for each participant and experimental session (i.
ventory (Oldfield, 1971), had normal or corrected to normal vision, and e., fNIRS of fMRI). Additionally, during the fNIRS session, a five-minute
were native English speakers. Informed consent was obtained for all period of baseline activity was collected before starting the functional
subjects prior to participation in the study as approved by the human tasks. During this baseline period participants viewed a blank (black)
subjects institutional review board at Medical College of Wisconsin. screen in a quiet room while in a restful wakeful state.

2.2. Experimental equipment 2.4. Data analysis

Functional NIRS data was collected using a continuous wave optical 2.4.1. fNIRS
system at 760 and 850 nm with a 25 mW maximum power output, and a All fNIRS channel-level data analysis was completed using MATLAB
sampling frequency of 3.125 Hz. All fNIRS data was collected using the R2020a (The Mathworks, Natick, MA), together with NIRS Brain
NIRScoutX system (NIRx Medical Technologies, Glen Head, NY) Analyzer Toolbox (Santosa et al., 2018). Prior to analysis, raw light
together with the NIRStar software (ver. 15–2). To capture full-head absorbance data were converted into optical densities. Artifacts related
data, 32 emitters and 32 detectors were placed about the scalp with to head movement during the experiment were removed using temporal
an inter-optode distance of approximately 3 cm for a total of 102 derivative distribution repair (TDDR) motion correction (Fishburn et al.,
channels (i.e., emitter-detector pairs). Subject specific channel locations 2019). Following motion correction, the optical densities of each
and anatomical landmarks were digitized using a Polhemus FastTrack channel were converted into separate measures of oxy- (HbO) and
system (Polhemus, Colchester, VT). During fNIRS testing, participants deoxyhemoglobin (HbR) using the Beer-Lambert Law (Jacques, 2013).

2
A. Zinos et al. NeuroImage 290 (2024) 120569

Scalp blood flow effects were removed using a whole head principal chromophore concentration across n voxels affect m fNIRS sensor level
component analysis (PCA) of the 5-minute baseline period (Santosa recordings,
et al., 2020; Zhang et al., 2005). PCA components containing
tm× 1 = Am×n xn× 1 + ε (4)
scalp-related signals were identified based on the frequency content and
spatial patterns of activity across the 102 channels for each hemoglobin where tm× 1 is a vector of the t-statistic across fNIRS channels obtained
species. Components containing a global (rather than focal) spatial based on the beta weights, xn× 1 are the changes in hemoglobin con­
distribution, as well as those whose frequency content matched physi­ centration across the n voxels of the anatomical volume, and ε is a
ological phenomena (heart rate, respiration, Mayer waves) were iden­ normally distributed noise term. The sensitivity matrix Am×n was
tified and removed. Typically, 1–3 components were removed from the computed in MCXLAB using an optical Monte Carlo approach (Fang and
HbO time series of each participant while 0–2 components were Boas, 2009; Yu et al., 2018). Skin, skull, cerebrospinal fluid (CSF), white
removed from HbR. and grey matter were automatically segmented using the CAT12 soft­
The first-level analysis was applied at the single subject level using a ware package (Gaser et al., 2022) and light scattering paths through the
general linear model (GLM) to identify task-related changes in HbO and tissues were simulated using 107 photons for each source-detector pair
HbR for each fNIRS channel. An autoregressive (AR) prewhitening filter to estimate the sensitivity for detecting chromophore changes in the
and iteratively reweighted least squares (IRLS) algorithm were used cortical volume. The estimated tissue optical properties used in the
during the GLM analysis (Barker et al., 2013), to remove serially Monte Carlo forward modeling are shown in Table 1, (Strangman et al.,
correlated signals unrelated to the task design. Specifically, the GLM 2003).
model was given by: Fig. 1 shows an example of the sensitivity matrix mapped to the
y = Xβ + ϵ (1) cortical surface for a single participant to illustrate the spatial coverage
of the NIRS sensor montage. After the sensitivity matrix was computed,
were y is the measured hemodynamic data, X is the design matrix the forward model (Eq. (4)) was inverted to solve for xn× 1 , which cor­
defined by the convolution of the canonical hemodynamic response responds to the effective light absorption at each voxel based on the
function with the block design structure of the tasks, and ϵ is assumed to normalized intensity measure (t-statistic) obtained for each optode
be normally distributed (~N(0, σ2)) noise. Model coefficients, β, were channel. A regularized least-squares approach, Low Resolution Elec­
initialized by inverting X via ordinary least squares method. The model tromagnetic Tomography (LORETA), was then used to solve the inverse
was then iteratively solved to remove serially correlated noise by fitting problem (Pascual-Marqui et al., 1994). LORETA uses a penalty term that
the model residual (y − Xβ) to an AR model. The resulting AR co­ includes the discrete Laplacian operator, L, to find the smoothest spatial
efficients were used to generate a prewhitening filter, F, that was then distribution ̂ x , that accounts for the channel-level measurements
applied to the hemodynamic data and design matrices such that, ( )
̂x = argminx ‖ t − Ax ‖2 + α‖ Lx ‖2 (5)
Fy = FXβ + Fϵ (2)
where A, x, t are defined according to Eq. (4), and α is a regularization
The process was repeated using IRLS until changes in β were less than
parameter optimized separately using the l-curve method (Hansen,
1 % to remove non-task related serially correlated signals from the GLM
1992).
fit.
Volumetric fNIRS sources estimated from LORETA were then
A second level analysis was applied at the group level using a mixed
thresholded to identify regions of significant activity. Since fMRI can
effects model to characterize the functional activity present across par­
detect signals in regions of the brain that fNIRS is known to be insen­
ticipants:
sitive too, the comparison between imaging modalities focused on
β = A⋅Γ + η (3) evaluating the correspondence (or lack thereof) in regions of the brain
that both fNIRS and fMRI were capable of sensing. For this reason, a
where β are the beta weights from the first level GLM analysis of each sensitivity-based threshold was applied to the LORETA results, masking
participant, A is the mixed effects model, Γ represents the task condi­ out voxels that fell below 1 % of the maximum total fNIRS sensitivity
tions, and η is a normally distributed error term to account for between (Habermehl et al., 2014). The same voxels were also masked from the
subject variability. Beta weights were evaluated for statistical signifi­ fMRI during comparison. Following the sensitivity-based threshold, an
cance using a t-test corrected for multiple comparisons across fNIRS amplitude threshold was applied to the fNIRS activity based on the
channels using the Benjamini Hochberg method (Benjamini and Hoch­ volumetric null distribution of fNIRS activity. The null distribution was
berg, 1995). estimated by assigning t-score normalized signal amplitudes to each of
the fNIRS channels drawn independently from a standard normal dis­
2.4.2. Source reconstruction tribution. A random sample drawn from the channel-level distribution
Channel-based analyses of fNIRS data contain limited spatial infor­ was then multiplied by the inverse model that solves Eq. (5) to obtain a
mation and cannot be compared to fMRI results directly. DOT source single sample of the volumetric null distribution. The sampling process
reconstruction techniques can be used to estimate the voxel-wise was repeated 10,000 times to accurately estimate the null distribution at
changes in chromophore concentration from the channel results, using each voxel. The amplitude threshold was then defined by the p-value
a “forward model” of the light absorption and scattering within the specifying statistical significance relative to the null distribution (e.g., p
anatomy. Source reconstruction of fNIRS activity was performed in
Brainstorm v. 3.220401 (Tadel et al., 2011), which is freely available
under the GNU general public license (http://neuroimage.usc.edu/b Table 1
rainstorm), using the Nirstorm plugin (Cai et al., 2021) and MCXLAB Estimated tissue optical properties used in optical Monte Carlo (760 nm/850
(Fang and Boas, 2009; Yu et al., 2018). A forward model was generated nm, both in mm− 1). Absorption and scattering coefficients are denoted by μa and
in Nirstorm for each participant based on their individual anatomical μs respectively, while ’g’ denotes the anisotropy factor and ’n’ the refraction
MRI scan and the digitized fNIRS optode locations obtained during index.
fNIRS data collection. Optode locations were registered to each partic­ μa μs g n
ipant’s anatomical MRI scan by aligning the nasion, and the left/right
Scalp 0.0177/0.0191 9.125/8.25 0.92 1.37
pre-auricular points, and were verified against the vitamin E capsule Skull 0.0125/0.0136 11.625/10.75 0.92 1.37
fiducials recorded by MRI. A sensitivity matrix (Amxn) was then CSF 0.0021/0.0026 0.125/0.125 0.92 1.37
computed for each participant, describing how perturbations in Brain (GM+WM) 0.0195/0.0186 14.75/13.875 0.92 1.37

3
A. Zinos et al. NeuroImage 290 (2024) 120569

Fig. 1. fNIRS sensor montage registered to the head for participant #2144 (upper-left). Emitters and detectors are colored red and green respectively. The remaining
graphics show the corresponding summed sensitivity map for the subject, illustrating the relative sensitivity of the fNIRS sensor montage (shown on a square-root
scale) for detecting chromophore changes across the cortical surface. Anterior (A), posterior (P), superior (S), inferior (I), left (L) and right (R) are labelled
for reference.

< 0.05). Since there was no a priori reason to assume an arbitrary analysis was performed with 3dttest++ using the contrasts of interest
p-value, such as 0.05, would be most appropriate for comparison with between the regressor parameter maps for each participant. Regions of
the fMRI results, multiple p-value thresholds were applied to each sub­ significant fMRI activity were defined using an amplitude threshold (p
ject and task over the range [0.0001, 0.1]. The threshold that resulted in < 0.001) and provided as input to 3dClustSim to determine the mini­
the highest mean Dice-Sorensen coefficient (DSC) across subjects was mum cluster size to achieve a corrected significance of p < 0.05.
identified separately for each task. This resulted in a within subject Group-level results were mapped to the Colin27 brain (Holmes et al.,
p-value threshold of 0.0599 for the Motor task, and 0.0405 for the Visual 1998) for visualization.
task.
Group level fNIRS channel results were projected to the Colin27 2.4.4. Comparison between fNIRS and fMRI
brain model (Holmes et al., 1998) using LORETA together with the The LORETA source reconstruction of fNIRS data was compared with
average sensor locations obtained by mapping participant-specific the first level fMRI results for each task, participant, and chromophore.
sensor locations onto the Colin27 head model and calculating the Since both fNIRS and fMRI image volumes were co-registered in the
spatial average for each sensor across participants. Thresholds for fNIRS same space, all comparisons were performed on a voxel-wise basis. fMRI
amplitude and sensitivity were applied using the approach detailed was treated as the gold standard such that voxels in which both fNIRS
above for the individual analyses. and fMRI were detected were labelled “true positives” (TP), voxels with
only fNIRS present were labelled “false positives” (FP), and voxels in
2.4.3. fMRI which only fMRI was detected were labelled “false negatives” (FN).
All fMRI pre-processing was performed using fMRIPrep 1.4.1 (Este­ Since the activity of interest is superficial due to the insensitivity of
ban et al., 2019), which is based on Nipype 1.2.0 (Gorgolewski et al., fNIRS to deeper structures, a volume to surface projection was used to
2011). Briefly, all anatomical scans were corrected for intensity aid visualization of the overlap between regions of significant fNIRS and
non-uniformity, skull stripped, and segmented. Brain surfaces were then fMRI activity. A volume to surface projection was performed using
reconstructed using recon-all (FreeSurfer 6.0.1) (Dale et al., 1999). MR Brainstorm to generate an interpolation matrix relating the vertices of
functional scans were corrected for susceptibility distortions using two the brain surface mesh (representing the participants cortical surface as
echo-planar imaging (EPI) with opposing phase-encoding directions. segmented by CAT12) to voxel locations in the anatomical MRI. The
fMRI volumes were then co-registered to the T1-weighted image, motion presence of a statistically significant task-related fNIRS/fMRI activity at
corrected, and smoothed with a 4 mm full width at half maximum each vertex on the cortical surface was determined based on the
(FWHM) gaussian kernel. neighborhood of voxels nearest the vertex. If more than half the of the
First level BOLD signal changes for each task were analyzed in AFNI voxels mapped to a vertex did not include significant activity, the vertex
20.2.19 (Cox, 1996) using a GLM with a restricted maximum likelihood was considered not to have significant activation for that imaging mo­
estimate model (3dREMLfit) to estimate temporal auto-correlations. dality. Vertices meeting the threshold of significant activation were
Stimulus regressors were created for each task condition by color coded to indicate whether significant activity at each location
convolving the block design with the hemodynamic response. Six mo­ included only fNIRS (orange), only fMRI (green), or both (red). A fourth
tion parameters (x, y, and z translations as well as pitch, roll, and yaw color (blue) was included to label voxels with significant fMRI activity
rotations) were included as covariates of no interest. Second level group that were not included in the comparison due to a lack of signal

4
A. Zinos et al. NeuroImage 290 (2024) 120569

sensitivity within the NIRS montage. An example volume to surface whether there was a significant difference in the average (between
projection and corresponding overlap map is illustrated in Fig. 2 for a subject) overlap of task activity for fNIRS compared to fMRI.
single representative participant.
The spatial correspondence between fNIRS and fMRI activity in 3. Results
volume was quantified using the positive predictive value (PPV) and
true positive rate (TPR), Out of the 22 participants enrolled in the fNIRS-fMRI study, 20 (38.9
± 15.1 years, 6 male) completed both imaging sessions. One participant
TP
PPV = (6) terminated the fMRI session early due to occipital head discomfort and a
TP + FP
second terminated the fNIRS session early due to discomfort from
TP several NIRS optodes contained in the wearable skullcap. Among the
TPR = (7) remaining 20 participants, two had poor fNIRS signal calibrations in
TP + FN
channels over the motor area and were excluded from the subsequent
In this context, positive predictive value indicates the proportion of
analysis. Within subject and group analyses were performed on the
fNIRS activity that corresponds spatially with fMRI, while TPR indicates
remaining 18 participants (40.5 ± 15.0 years, 4 male) for the finger
the proportion of fMRI activity that is successfully localized by fNIRS.
tapping task, and on a subset of 9 participants (37.5 ± 14.3 years, 2
Due to the focal nature of the activity elicited by the Motor and Visual
male) who also performed the Visual task.
tasks, the specificity (or true negative rate) was high since most voxels
were classified as correct rejections (no signal in either modality). As a
3.1. fNIRS channel results
result, specificity was not considered a meaningful metric for the current
comparison. Instead, the DSC (Sorensen, 1948; Dice, 1945) was calcu­
Fig. 3 shows task-related activity across fNIRS channels for the Motor
lated for each comparison, consistent with its use as a similarity metric
task of a typical participant (#2144). Significant increases in HbO and
for characterizing binary image segmentation in MRI (Zou et al., 2004;
decreases in HbR are present in the contralateral motor cortex, along
Zijdenbos et al., 1994)
with similar, but smaller, signal changes in ipsilateral motor cortex.
2TP Significant activity is also present on occipital channels for both chro­
DSC = (8)
2TP + FP + FN mophores and task conditions, and in frontal channels for left-hand
The spatial variability in task-related activity for each imaging mo­ deoxyhemoglobin, but with the opposite sign. Block-averaged task re­
dality was quantified by calculating the distribution of DSC values be­ sponses on the channels with the strongest response for left and right-
tween all subject pairs. DSC values were calculated separately for the hand finger tapping are shown in the plot insets. Equivalent plots for
thresholded fMRI activity and the thresholded fNIRS HbO and HbR additional subjects and for the Visual task can be found in the supple­
source maps for the motor task (n = 18, giving 153 possible subject mentary materials.
pairs) but not for the visual task due to the limited sample size (n = 9).
Prior to calculating the DSC, each subject’s anatomy was normalized to 3.2. Individual comparison of fNIRS and fMRI
the MNI-ICBM152 brain template (Mazziota et al., 1995, 2001) via
SPM12 (update revision number 7771) (Penny et al., 2006), and the Surface maps of the overlap between the co-registered fMRI response
resulting deformation fields were applied to the co-registered functional and source localized fNIRS response for the Motor task are shown in
images. The resulting distributions of DSC values were then examined Fig. 4 for three representative participants (#2014, #2132, #2144)
and tested for significance using a Wilcoxon rank-sum test, to determine drawn from the first, second, and third quartiles of fNIRS/fMRI overlap
across participants. Similar results are shown for both chromophore

Fig. 2. Volume to surface projection of task-related activity measured during right-handed finger tapping for a representative participant (#2144). fNIRS results are
shown here for oxyhemoglobin. Each slice image shows the volumetric activity from a different view (axial, sagittal and coronal) centered on motor cortex (MNI
coordinates x: − 40.72, y: − 22.70, z: 60.54), together with the corresponding surface projection. Anterior (A), posterior (P), left (L) and right (R) are labelled
for reference.

5
A. Zinos et al. NeuroImage 290 (2024) 120569

Fig. 3. Single subject (#2144) fNIRS channel level results for the Motor task. Each topographic plot shows the T-score for task-related activity in each channel.
Channels with significant task-related are denoted by solid-colored lines (p < 0.05, critical T = 2.47 via Benjamini-Hochberg procedure). Channels whose task-
related activity was not statistically significant are denoted by dashed lines. Plot insets show the block average responses and standard deviations during the task
period (green shading) for the channel with the largest task-related response. Time courses have been smoothed by a 3-second centered moving average (mean) for
visualization.

Fig. 4. Single subject comparisons of the spatial correspondence between task-related fNIRS and fMRI activity mapped to the cortical surface during the Motor task
for three representative participants drawn from the first, second, and third quartiles of fNIRS/fMRI overlap (#2014, #2132, #2144, labelled as I, II, III respectively).
For all comparisons, the fNIRS sensitivity threshold was set to 1 % of the maximum total sensitivity and the amplitude threshold (p-value) was set to the average
values across participants (p < 0.0599) that maximized the spatial correspondence with fMRI during the Motor task. Anterior (A), posterior (P), left (L) and right (R)
are labelled for reference.

6
A. Zinos et al. NeuroImage 290 (2024) 120569

species. The primary region of overlap between fNIRS and fMRI 0.47 for HbO, and 0.48 for HbR. The difference in median DSC between
occurred in the contralateral somatosensory/motor cortices for all par­ the hemoglobin chromophores was not significant (two-tailed Wilcoxon
ticipants, which corresponded to the region of highest activity in the ranked-sum test, p = 0.44). However, the difference in median DSC
channel level analysis. In one of the participants shown (#2144), fMRI between fMRI and the fNIRS hemoglobin chromophores was significant
and fNIRS both measured significant activity in the ipsilateral motor (two-tailed Wilcoxon ranked-sum test, p < 0.05) for both HbO and HbR.
cortex, particularly during left-handed finger tapping. Clusters of false The distributions and medians of the DSC for all subject pairs are shown
positive (fNIRS but not fMRI) activity are present in each comparison. for each imaging modality in the supplemental materials (Figure S6).
Regions of fMRI-only activity were generally adjacent to fNIRS/fMRI
active regions except for a cluster in the ipsilateral parietal cortex for 3.3. Group level comparison of fNIRS and fMRI
participant #2144 (panel III, right) during the Right Motor task. Sig­
nificant sulcal and interhemispheric activations that were present in the Fig. 6 shows surface maps of the overlap between the group level
fMRI maps fell below the fNIRS sensitivity threshold, indicating regions fMRI response and source localized fNIRS response during the finger
of task-related activity that fNIRS was not sensitive to. tapping tasks. Both modalities detected significant task-related activity
Fig. 5 shows the fNIRS/fMRI comparison of activity during the Visual in the contralateral somatosensory and motor cortices. The group level
task for the same three representative participants as the Motor task. For fNIRS activity is sparser, resulting in a higher PPV and improved overall
both fNIRS and fMRI, task-related activity was centered around occipital DSC compared to within-subject analyses (Table 3). Small clusters of
cortex. Functional MRI only activity tended to be located more posterior task-related fMRI activity not matched by fNIRS were present, particu­
and inferior to the fNIRS clusters, and while these areas did not fall larly in ipsilateral motor cortex, resulting in a similar TPR to the within-
below the sensitivity threshold selected for this analysis it should be subject analyses. The visual task was not analyzed at the group level
noted that fNIRS sensitivity tended to be lower in these regions. False because of the relatively small number of subjects that completed the
positive clusters of fNIRS activation were sparser than for the finger task.
tapping task and primarily located in regions adjacent to fMRI clusters of
activity. TPR, PPV and DSC all varied with task, and average values for 4. Discussion
each task and chromophore are reported in Table 2 along with the range
of values across subjects. Significant variations were noted across The aim of this work was to assess the correspondence between
subjects. source-localized fNIRS and BOLD fMRI, with the goal of better under­
The spatial variability of fMRI motor activity between subjects, standing how fNIRS can be used clinically. Because both fNIRS and
quantified as the median pairwise DSC measured between all subject BOLD fMRI ultimately measure the hemodynamic response, it was ex­
pairs, was 0.45. The median pairwise DSC for fNIRS motor activity was pected there would be significant concordance between modalities.

Fig. 5. Single subject comparisons of the of the spatial correspondence between task-related fNIRS and fMRI activity mapped to the cortical surface during the Visual
task for three representative participants drawn from the first, second, and third quartiles of fNIRS/fMRI overlap (#2014, #2132, #2144, labelled as I, II, III
respectively). For all comparisons, the fNIRS sensitivity threshold was set to 1 % of the maximum total sensitivity and the amplitude threshold (p-value) was set to the
average values across participants (p < 0.0405) that maximized the spatial correspondence with fMRI during the Visual task. Anterior (A), posterior (P), left (L) and
right (R) are labelled for reference.

7
A. Zinos et al. NeuroImage 290 (2024) 120569

Table 2
Mean, minimum and maximum TPR, PPV, and DSC of the volumetric spatial correspondence between task-related fNIRS/fMRI activity within-subject across all
participants. HbO indicates oxyhemoglobin and HbR indicates deoxyhemoglobin.
TPR PPV DSC

n Hb Mean Min./Max. Mean Min./Max. Mean Min./Max.

Right Motor 18 HbO 0.50 [0.02, 0.88] 0.40 [0.06, 1.00] 0.36 [0.03, 0.66]
HbR 0.47 [0.05, 0.85] 0.41 [0.03, 0.98] 0.36 [0.04, 0.64]
Left Motor 18 HbO 0.47 [0.14, 0.93] 0.43 [0.12, 0.83] 0.40 [0.16, 0.64]
HbR 0.45 [0.06, 0.81] 0.42 [0.02, 0.79] 0.39 [0.03, 0.59]
Visual 9 HbO 0.52 [0.07, 0.87] 0.55 [0.31, 0.75] 0.47 [0.14, 0.61]
HbR 0.50 [0.20, 0.90] 0.45 [0.17, 0.93] 0.39 [0.24, 0.54]

Fig. 6. Group level comparison of the spatial correspondence between significant task-related fNIRS and fMRI activity mapped to the cortical surface during the
Motor task. fNIRS sensitivity threshold was set to 1 % of the maximum total sensitivity and the group amplitude threshold that maximized the correspondence
between fNIRS and fMRI was (p < 0.0003). Anterior (A), posterior (P), left (L) and right (R) are labelled for reference.

to the use of the LORETA algorithm during source reconstruction which


Table 3
has a smoothness-based regularization term that favors broader activa­
TPR, PPV, and DSC of the group-level spatial correspondence between volu­
tions over focal ones. While LORETA can improve the detection of
metric task-related fNIRS/fMRI activity. HbO indicates oxyhemoglobin and HbR
indicates deoxyhemoglobin.
sources in depth (Tremblay et al., 2018), it can also reduce the sensi­
tivity to more focal sources (Pascual-Marqui et al., 1994). For example,
Hb TPR PPV DSC
while contralateral somatosensory and primary motor areas were
Left Motor HbO 0.68 0.50 0.58 consistently detected at the group level with both fNIRS and fMRI,
HbR 0.29 0.50 0.37
ipsilateral motor activity was detected primarily with fMRI. Bilateral
Right Motor HbO 0.57 0.41 0.48
HbR 0.58 0.63 0.60
motor activity during a unilateral motor task is well documented with
fMRI (Kobayashi et al. 2003; Beltramello et al. 1998; Cramer et al. 1999,
Kim et al., 1993), and was observed here with fNIRS in a subset of
Group level results were calculated for comparison with previous work, participants. More robust group level detection of focal sources with
however, the primary interest was in characterizing the detection of fNIRS could be realized by increasing the density of the whole-head
brain activity within-subject, which is ultimately needed to maximize sensor montage and/or by using a spatially heterogenous montage tar­
the clinical utility of fNIRS. getted to areas of low relative fNIRS sensitivity.
At the group level, fNIRS corresponded well with fMRI activation, Functional NIRS analysis within subjects was nearly as sensitive to
with a DSC ranging from 0.37 to 0.6. These results are consistent with fMRI as in the group level comparison with an overall average TPR of
previous work showing significant correlation between fNIRS and fMRI 47.25 % for Motor tasks and 51 % for the Visual task. The ability to
at the group level (Huppert et al., 2017; Wijeakumar et al., 2017). maintain the overall TPR within-subject is associated with improved
Additionally, TPR ranged from 29 % to 68 % (average: 53 %). This spatial registration of fNIRS activity to cortex vis-à-vis the 3D measure­
compares well with the 59 % maximum reported by Hernández-Martin ment of optode locations and subsequent registration onto the partici­
and Hernández-Martin (2017), during a cognitive arithmetic problem pant’s specific head model. Single subject (vs. group). fNIRS was
task, and exceeds the 26.2 % average TPR obtained across all study tasks somewhat less specific in the Motor task, reflecting an increase in false
(Hernández-Martin et al., 2017). No clear advantage was noted in using positive activation (relative to fMRI) present within subject. The activity
HbO versus HbR as a better match to fMRI results, and existing literature could reflect the presence of scalp blood flow or motion artifacts that
has found conflicting results with some (Cui et al., 2011; Strangman were not fully removed from the within-subject fNIRS analysis. Inter­
et al., 2002) reporting better correspondence in HbO and others estingly, spatial localization of the Visual task was more consistent than
(Kleinschmidt et al., 1996) better correspondence in HbR. One notable the Motor task. In spite of the similar performance scores, the spatial
difference relative to previous results was the similarity in spatial extent distribution of false positive fNIRS activity was consistently clustered
obtained between fNIRS and fMRI activity. Previous group comparisons around the fMRI activity in the Visual task suggesting a primary impact
have tended to yield more focal representations of fNIRS relative to fMRI of spatial resolution with the whole-head montage rather than the
(Huppert et al., 2017; Wijeakumar et al., 2017) This could be due in part presence of spurious signals. When comparing the results of the Visual

8
A. Zinos et al. NeuroImage 290 (2024) 120569

task with those of the Motor task, one limitation of this research is the methodology between the two imaging modalities. In the fMRI analysis,
difference in sample size, with only 9 participants completing the Visual the amplitude map was first thresholded to attain a voxel-wise signifi­
task compared to 18 for the Motor task. This limits the statistical power cance of p < 0.001. Supra-threshold values were then cluster-extent
of the Visual group results, although individual comparisons are thresholded, a secondary process based on gaussian random field the­
unaffected. ory that implements a form of multiple comparisons correction by
Overall, individual results were variable, with DSCs ranging from a masking out small clusters of voxels while maintaining large clusters
low of 3 % (slight to no overlap) to a high of 66 %, (moderate to sub­ unlikely to arise due to random chance (Woo et al., 2014). In fNIRS,
stantial) (Wilson et al., 2016). The wide range could reflect individual amplitude thresholding was also used to retain only the most significant
differences in the relative positions of active cortical regions with values, but cluster-extent thresholding was not applied, as it was unclear
respect to the fNIRS optodes, and thus spatial differences in relative whether the assumptions of gaussian random field theory would hold for
sensitivity to changes in HbO and HbR. Individual results could also be the reconstructed volumetric fNIRS data. A consequence of this differ­
particularly sensitive to correct tissue segmentation. While it was ence is that fNIRS data may display more false positives with respect to
beyond the scope of this paper, the robustness of fNIRS reconstruction to fMRI. Adapting cluster-extent thresholding to fNIRS could lead to fewer
correct tissue segmentations and tissue optical properties has been dis­ false positive activations and improve fNIRS/fMRI overlap, particularly
cussed in previous literature (Strangman et al., 2003; Strangman et al., for within subject analyses.
2013). During source reconstruction the spatial variation in fNIRS sen­ The results of this study support the feasibility of using fNIRS clini­
stivity across cortex can impact the extent of fNIRS activity even when cally to characterize brain activity in individual subjects, particularly in
fNIRS and fMRI are in agreement. This result is not unexpected since the situations where information about regional brain activity is needed. For
point spread function of fNIRS with a sparse (3 cm channel) square array example, fNIRS has been shown to be sensitive to the changes in blood
has been reported to have a FWHM of 21.46 ± 4.6 mm, (White and oxy- and deoxygenation that occur during periods of abnormal electrical
Culver, 2010). By comparison, the FWHM of activity measured with 3T activity in people diagnosed with epilepsy (Monrad et al., 2015). During
gradient echo fMRI is 3.9 ± 0.7 mm (Parkes et al., 2005). Moreover, a seizure, there is a decrease in oxygenation in the brain due to increased
fNIRS analyses may be sensitive to other sources of task correlated ac­ neural activity. By measuring these changes, whole head fNIRS (i.e.,
tivity (signal or noise) that are not reflected in fMRI BOLD analyses. DOT), can provide valuable insight into the activity and progression of a
fNIRS is known to be sensitive to scalp blood flow and motion artifacts seizure, allowing for more precise diagnosis and treatment. Whole head
(Huppert, 2016), which can become task locked and create false positive fNIRS could be particularly useful in the diagnosis of focal seizures,
activations that are not cortical in origin. The incorporation of short which originate from a specific area of the brain. The integration of
channel detectors (<1 cm separation) into the fNIRS montage can be fNIRS with EEG could improve noninvasive monitoring and localization
used to directly measure non-cortical contributions to the fNIRS signal of seizure activity, enabling more precise diagnosis and treatment of
and could further improve the correspondence between fNIRS and fMRI focal seizures. fNIRS may also be capable of assessing Cytochrome Ox­
within subject (Saager and Berger, 2008). idase REDOX states in the brains of epilepsy patients and could shed
The spatial variability of fNIRS and fMRI DSC values between sub­ light on the metabolic and hemodynamic change that occur during
jects was comparable, however, there was a small but significant dif­ seizure, such as uncoupling of oxidative phosphorylation, or alterations
ference between fMRI (DSC = 0.45), and fNIRS-HbO (DSC = 0.47) and in cerebrovascular autoregulation.
fNIRS-HbR (DSC = 0.48) respectively. The moderate DSC values for both Future work developing fNIRS as a clinical tool should focus on
fMRI and fNIRS indicate some individual differences in task-related characterizing the sources of individual variability, and determining the
activation, even after normalization to a template brain. The reduced extent to which fNIRS signals are repeatable within subject. To the
spatial variability observed with fNIRS may reflect blurring coupled extent that fNIRS-only regions of activity are not cortical in origin, (e.g.,
with lower spatial resolution, which would tend to produce more similar scalp blood flow, Mayer waves, etc.), more effective sensing and/or
spatial maps from one subject to another. signal processing methods may be required to increase clinical effec­
With an average DSC of 38 % in the Motor task and 43 % in the Visual tiveness. While a baseline PCA method was used in this work to remove
task, the individual correspondence between fNIRS and fMRI was scalp bloodflow and noncortical sources of activity (Santosa et al., 2020;
comparable to the group level analysis reported here and in other studies Zhang et al., 2005), numerous other methods have been proposed
(Hernández-Martin et al., 2017). Although PPV was limited, the 49 % including independent component analysis (Zhao et al., 2021), wavelet
average TPR across participants indicates that fNIRS detected approxi­ coherence based filtering (Duan et al., 2018), Kalman filtering (Orte­
mately half of active fMRI voxels. Focal sensor placement around re­ ga-Martinez et al., 2022), and the use of short seperation channels
gions of interest (Anwar et al., 2016; Strangman et al., 2002; Toronov (Saager and Berger, 2008). Another issue that could arise in clinical
et al., 2001; Hernández-Martin et al., 2017; Klein et al., 2022; Wijea­ applications is the need for online algorithms capable of real-time
kumar et al., 2017; Huppert et al., 2006; Huppert et al., 2017), could be monitoring. Real-time monitoring requires that fNIRS be validated
used to increase sensor density locally and improve the correspondence both spatially and temporally. Though fNIRS has been repeatedly shown
between fNIRS and fMRI. Spatially localized high-density montages such to have strong temporal correlations to fMRI (Cui et al., 2011; Noah
as the personalized optimal montage approach (Machado et al., 2014; et al., 2015; G. Strangman et al., 2002) further work is needed to show
Cai et al., 2021) can result in an improved signal to noise ratio which can that the temporal correspondence is sufficient for real-time clinical
allow lower amplitude signals to be detected. Such approaches could monitoring. This use case also requires computationally efficient algo­
increase detection of fNIRS signals from deeper regions of cortex that rithms, and specialized statistical methods. The process of projecting
have low sensitivity with a whole-head montage. The use of a high fNIRS channel data onto a cortical surface map can be computationally
density grid of fNIRS optodes has been shown to improve uniformity of intensive due to the need to invert the forward sensitivity model. Future
the sensitivity profile and reduce localization error by up to a factor of work could assess the practicality of inverting the forward model once,
five (White and Culver, 2010). Higher density arrays of fNIRS optodes on a set of calibration data, and then using that inverse model for all
may be required if attempting to detect a task response from regions that future time points. This would reduce the computational complexity of
are not superficial, such as in sulci. Here, however, whole head coverage the fNIRS cortical projection to a series of matrix-vector multiplications,
was used with an eye toward generalized clinical applications within which could be more easily performed in real-time in a clinical setting.
subject in which all regions of relevant activity may not be known a
priori. 5. Conclusion
In addition to sensitivity to scalp blood flow and motion artifacts, the
PPV may have been impacted by differences in the statistical Whole head fNIRS could provide a valuable bedside neuroimaging

9
A. Zinos et al. NeuroImage 290 (2024) 120569

tool in various clinical applications if it can be shown to be reliable and References


accurate in individual patients (versus group analysis). The results of
this study show a strong correspondence at the group level between Anwar, A.R., et al., 2016. Effective connectivity of cortical sensorimotor networks during
finger movement tasks: a simultaneous fNIRS, fMRI, EEG Study. Brain Topogr. 29
fNIRS and fMRI task-related changes in hemodynamic brain activity. At (5), 645–660. https://doi.org/10.1007/s10548-016-0507-1.
the level of individual participants, the pre-processing steps coupled Barker, J.W., Aarabi, A., Huppert, T.J., 2013. Autoregressive model based algorithm for
with the incorporation of subject-specific anatomy and sensor position correcting motion and serially correlated errors in fNIRS. J. Biomedical Optics
Express 4 (8), 1366–1379.
information increased the sensitivity of fNIRS for detecting and local­ Beltramello, A., et al., 1998. Motor representation of the hand in the human cortex: an f-
izing regions of significant fMRI activity. A subset of participants yielded MRI study with a conventional 1.5 T clinical unit. Ital. J. Neurol. Sci. 19 (5),
a relatively high number of “false positive” regions where fNIRS 277–284. https://doi.org/10.1007/BF00713853.
Benjamini, Y., Hochberg, Y., 1995. Controlling the False Discovery Rate: a Practical and
measured activity not found by fMRI. These individual differences likely Powerful Approach to Multiple Testing. Series B Methodolog. 57 (1), 289–300.
reflect disproportionately higher contributions of non-cortical sources of Cai, Z., et al., 2021. Evaluation of a personalized functional near infra-red optical
near infrared absorbance (e.g., heterogenous changes in scalp blood tomography workflow using maximum entropy on the mean. Hum. Brain Mapp. 42
(15), 4823–4843. https://doi.org/10.1002/hbm.25566.
flow, Mayer waves, etc.), that have not been fully accounted for during
Chen, W., et al., 2020. Functional Near-Infrared Spectroscopy and Its Clinical Application
the analysis. The incorporation of short-distance detectors to directly in the Field of Neuroscience : advances and Future Directions. Front. Neurosci. 14
measure non-cortical contributions could further improve within- (July), 1–15. https://doi.org/10.3389/fnins.2020.00724.
subject detection of cortical activity and the clinical utility of whole- Cox, R.W., 1996. AFNI: software for analysis and visualization of functional magnetic
resonance neuroimages. Comp. Biomed. Res. 29, 162–173.
head fNIRS for bedside monitoring of brain activity. Cramer, S.C., et al., 1999. Activation of distinct motor cortex regions during ipsilateral
and contralateral finger movements. J. Neurophysiol. 81 (1), 383–387. https://doi.
CRediT authorship contribution statement org/10.1152/jn.1999.81.1.383.
Cui, X., Bray, S., Bryant, D.M., Glover, G.H., Reiss, A.L., 2011. A quantitative comparison
of NIRS and fMRI across multiple cognitive tasks. Neuroimage 54 (4), 2808–2821.
Anthony Zinos: Data curation, Formal analysis, Methodology, https://doi.org/10.1016/j.neuroimage.2010.10.069.
Software, Visualization, Writing – original draft, Writing – review & Dale, A.M., Fischl, B., Sereno, M.I., 1999. Cortical surface-based analysis I. Segmentation
and surface reconstruction. Neuroimage (9), 179–194.
editing. Julie C. Wagner: Data curation, Formal analysis, Methodology, DeYoe, E.A., Ulmer, J.L., Mueller, W.M., Sabsevitz, D.S., Reitsma, D.C., Pillai, J.J., 2015.
Software, Visualization, Writing – original draft, Writing – review & Imaging of the functional and dysfunctional visual system. Semin. Ultrasound CT MR
editing. Scott A. Beardsley: Conceptualization, Formal analysis, 36 (3), 234–248. https://doi.org/10.1053/j.sult.2015.05.015.
Dice, L.R., 1945. Measures of the amount of ecologic association between species.
Investigation, Methodology, Project administration, Resources, Soft­ Ecology. 26 (3), 297–302. https://doi.org/10.2307/1932409.
ware, Supervision, Validation, Writing – original draft, Writing – review Donizete de Faria, D., et al., 2020. Task-related brain activity and functional connectivity
& editing. Wei-Liang Chen: Conceptualization, Investigation, Valida­ in upper limb dystonia: a functional magnetic resonance imaging (fMRI) and
functional near-infrared spectroscopy (fNIRS) study. Neurophotonics. 7 (4), 1–13.
tion, Writing – original draft, Writing – review & editing. Lisa Conant:
https://doi.org/10.1117/1.NPh.7.4.045004.
Conceptualization, Investigation, Validation, Writing – original draft. Duan, L., et al., 2018. Wavelet-based method for removing global physiological noise in
Marsha Malloy: Conceptualization, Data curation, Investigation, Proj­ functional near-infrared spectroscopy. Biomed. Opt. Express. 9 (8), 3805–3820.
ect administration, Writing – original draft, Writing – review & editing. https://doi.org/10.1364/BOE.9.003805.
Esteban, O., et al., 2019. fMRIPrep: a robust preprocessing pipeline for functional MRI.
Joseph Heffernan: Methodology, Writing – original draft, Writing – Nat. Methods 16, 111–116. https://doi.org/10.1038/s41592-018-0235-4. Springer
review & editing. Brendan Quirk: Conceptualization, Project adminis­ US.
tration, Validation, Writing – original draft, Writing – review & editing. Fang, Q., Boas, D.A., 2009. Monte Carlo simulation of photon migration in 3D turbid
media accelerated by graphics processing units. Opt. Express 17, 20178–20190.
Robert Prost: Conceptualization, Formal analysis, Methodology, Fishburn, F.A., et al., 2019. Temporal Derivative Distribution Repair (TDDR): a motion
Writing – original draft, Writing – review & editing. Mohit Mahesh­ correction method for fNIRS. Neuroimage 184, 171–179. https://doi.org/10.1016/j.
wari: Conceptualization, Data curation, Writing – original draft, Writing neuroimage.2018.09.025. Elsevier Ltd.
Gaser, C. et al. (2022) Alzheimer’s disease neuroimaging initiative. "A computational
– review & editing. Jeffrey Sugar: Methodology, Writing – original anatomy toolbox for the analysis of structural MRI data", bioRxiv.
draft, Writing – review & editing. Harry T. Whelan: Conceptualization, Fox, P.T., Raichle, M.E., 1984. Stimulus rate dependence of regional cerebral blood flow
Funding acquisition, Investigation, Methodology, Project administra­ in human striate cortex, demonstrated by positron emission tomography.
J. Neurophysiol. 51 (5), 1109–1120. https://doi.org/10.1152/jn.1984.51.5.1109.
tion, Resources, Supervision, Validation, Writing – original draft, Gorgolewski, K., et al., 2011. Nipype: a flexible, lightweight and extensible neuroimaging
Writing – review & editing. data processing framework in Python. Front. Neuroinform. 5 (August), 1–15.
https://doi.org/10.3389/fninf.2011.00013.
Habermehl, C., et al., 2014. Optimizing the regularization for image reconstruction of
Declaration of competing interest cerebral diffuse optical tomography. J. Biomed. Opt. 19 (9), 96006. https://doi.org/
10.1117/1.JBO.19.9.096006.
There are no competing interests to disclose. Hansen, P.C., 1992. Analysis of discrete Ill-posed problems by means of the L-curve.
SIAM Review 34 (4), 561–580. http://www.jstor.org/stable/2132628.
Hernández-Martin, E., et al., 2017. Comparing diffuse optical tomography and functional
Data availability magnetic resonance imaging signals during a cognitive task: pilot study.
Neurophotonics. 4 (1), 1–15. https://doi.org/10.1117/1.NPh.4.1.
Please contact the corresponding author for availability of deiden­ Holmes, C.J., et al., 1998. Enhancement of MR images using registration for signal
averaging. J. Comput. Assist. Tomogr. 22 (2), 324–333.
tified clinical data sets. Data sets normally require Institutional Review Hong, K.S., Yaqub, M.A., 2019. Applilcation of functional near-infrared spectroscopy in
Board approval and data sharing agreements. the healthcare industry: a review. J. Innov. Opt. Health Sci. 12 (6), 1–91.
Huppert, T.J., et al., 2006. A temporal comparison of BOLD, ASL, and NIRS
hemodynamic responses to motor stimuli in adult humans. Neuroimage 29 (2),
368–382. https://doi.org/10.1016/j.neuroimage.2005.08.065.
Acknowledgements Huppert, T.J., et al., 2017. Comparison of group-level, source localized activity for
simultaneous functional near-infrared spectroscopy- magnetoencephalography and
simultaneous fNIRS-fMRI during parametric median nerve stimulation. Neurop 4
This work was supported by the Bleser Endowed Chair in Neurology (1), 1–13. https://doi.org/10.1117/1.NPh.4.1.015001.
administered through the Children’s Wisconsin Foundation (to HW), the Huppert, T.J., 2016. Commentary on the statistical properties of noise and its implication
Baumann Research Endowment (to HW) and Office of Naval Research on general linear models in functional near-infrared spectroscopy. Neurophotonics. 3
(1), 1–10.
Grant No: N00014–19–1–2560 (to HW). Jacques, S.L., 2013. Optical properties of biological tissues: a review. Phys. Med. Biology
58, R37–R61.
Supplementary materials Jobsis, F.F., 1977. Noninvasive, infrared monitoring of cerebal and myocardial oxygen
sufficiency and circulatory parameters. Science (1979) 198 (4323), 1264–1267.
Kim, S.G., Ashe, J., Hendrich, K., Ellermann, J.M., Merkle, H., Uğurbil, K.,
Supplementary material associated with this article can be found, in Georgopoulos, A.P., 1993. Functional magnetic resonance imaging of motor cortex:
the online version, at doi:10.1016/j.neuroimage.2024.120569.

10
A. Zinos et al. NeuroImage 290 (2024) 120569

hemispheric asymmetry and handedness. Science 261 (5121), 615–617. https://doi. Santosa, H., et al., 2020. Quantitative comparison of correction techniques for removing
org/10.1126/science.8342027. systemic physiological signal in functional near-infrared spectroscopy studies.
Klein, F., Debener, S., Witt, K., Kranczioch, C., 2022. fMRI-based validation of Neurophotonics. https://doi.org/10.1117/1.NPh.7.3.035009.
continuous-wave fNIRS of supplementary motor area activation during motor Santosa, H., et al., 2018. The NIRS Brain AnalyzIR toolbox. Algorithms. 11 (73), 1–33.
execution and motor imagery. Scientific Reports | 12, 3570. https://doi.org/ https://doi.org/10.3390/A11050073.
10.1038/s41598-022-06519-7. Sorensen, T., 1948. A method of establishing groups of equal amplitude in plant
Kleinschmidt, A., et al., 1996. Simultaneous recording of cerebral blood oxygenation sociology based on similarity of species and its application to analyses of the
changes during human brain activation by magnetic resonance imaging and near- vegetation on danish commons. Kongelige Danske Videnskabernes Selskab 5, 1–34.
infrared spectroscopy. J. Cerebral blood flow. Metabolism 16 (5), 817–826. https:// Strangman, G., et al., 2002. A quantitative comparison of simultaneous BOLD fMRI and
doi.org/10.1097/00004647-199609000-00006. NIRS recordings during functional brain activation. Neuroimage 17 (2), 719–731.
Kobayashi, M., et al., 2003. Ipsilateral motor cortex activation on functional magnetic Strangman, G., Franceschini, M.A., Boas, D.A., 2003. Factors affecting the accuracy of
resonance imaging during unilateral hand movements is related to interhemispheric near-infrared spectroscopy concentration calculations for focal changes in
interactions. Neuroimage 20 (4), 2259–2270. https://doi.org/10.1016/s1053-8119 oxygenation parameters. Neuroimage 18 (4), 865–879. https://doi.org/10.1016/
(03)00220-9. S1053-8119(03)00021-1.
Mazziotta, J.C., et al., 1995. A probabilistic atlas of the human brain: theory and Strangman, G.E., Li, Z., Zhang, Q., 2013. Depth sensitivity and source-detector
rationale for its development. The International Consortium for Brain Mapping separations for near infrared spectroscopy based on the colin27 brain template.
(ICBM). Neuroimage 2 (2), 89–101. https://doi.org/10.1006/nimg.1995.1012. PLoS. One 8 (8), 66319. https://doi.org/10.1371/journal.pone.0066319.
Mazziotta, J., et al., 2001. A four-dimensional probabilistic atlas of the human brain. Tadel, F., Baillet, S., Mosher, J.C., Pantazis, D., Leahy, R.M., 2011. Brainstorm: a user-
JAMIA 8 (5), 401–430. https://doi.org/10.1136/jamia.2001.0080401. friendly application for MEG/EEG analysis. Comput. Intell. Neurosci. 2011, 879716.
Machado, A., et al., 2014. Optimal optode montage on electroencephalography/ https://doi.org/10.1155/2011/879716.
functional near-infrared spectroscopy caps dedicated to study epileptic discharges. Toronov, V., Webb, A., Choi, J.H., Wolf, M., Michalos, A., Gratton, E., Hueber, D., 2001.
J. Biomed. Opt. 19 (2), 026010 https://doi.org/10.1117/1.JBO.19.2.026010, 2014 Investigation of human brain hemodynamics by simultaneous near-infrared
FebPMID: 24525860. spectroscopy and functional magnetic resonance imaging. Med. Phys. 28 (4),
Monrad, P., et al., 2015. Haemodynamic response associated with both ictal and 521–527. https://doi.org/10.1118/1.1354627.
interictal epileptiform activity using simultaneous video electroencephalography/ Tremblay, J., et al., 2018. Comparison of source localization techniques in diffuse optical
near infrared spectroscopy in a within-subject study. J. Near. Infrared. Spectrosc. 23 tomography for fNIRS application using a realistic head model. Biomed. Opt.
(4), 209–218. https://doi.org/10.1255/jnirs.1170. Express. 9 (7), 2994–3016. https://doi.org/10.1364/BOE.9.002994.
Noah, J.A., Ono, Y., Nomoto, Y., Shimada, S., Tachibana, A., Zhang, X., Bronner, S., White, B.R., Culver, J.P., 2010. Quantitative evaluation of high-density diffuse optical
Hirsch, J., 2015. fMRI validation of fNIRS measurements during a naturalistic task. tomography: in vivo resolution and mapping performance. J. Biomed. Opt. 15 (2),
J. Vis. Exp. (100), e52116 https://doi.org/10.3791/52116. 026006 https://doi.org/10.1117/1.3368999.
Obrig, H., 2014. NIRS in clinical neurology - a “promising” tool? Neuroimage 85, Wijeakumar, S., et al., 2017. Validating an image-based fNIRS approach with fMRI and a
535–546. working memory task. Neuroimage 147, 204–218.
Oldfield, R.C., 1971. The assessment and analysis of handedness: the Edinburgh Wilson, S.M., Bautista, A., Yen, M., Lauderdale, S., Eriksson, D.K., 2016. Validity and
inventory. Neuropsychologia 9, 97–113. reliability of four language mapping paradigms. Neuroimage Clin. 16, 399–408.
Ortega-Martinez, A., et al., 2022. Multivariate Kalman filter regression of confounding https://doi.org/10.1016/j.nicl.2016.03.015.
physiological signals for real-time classification of fNIRS data. Neurophotonics. 9 Woo, C.W., Krishnan, A., Wager, T.D., 2014. Cluster-extent based thresholding in fMRI
(2), 025003 https://doi.org/10.1117/1.NPh.9.2.025003. analyses: pitfalls and recommendations. Neuroimage 91, 412–419. https://doi.org/
Parkes, L.M., et al., 2005. Quantifying the spatial resolution of the gradient echo and spin 10.1016/j.neuroimage.2013.12.058.
echo BOLD response at 3 Tesla. Magn. Reson. Med. 54, 1465–1472. https://doi.org/ Yu, L., et al., 2018. Scalable and massively parallel Monte Carlo photon transport
10.1002/mrm.20712. simulations for heterogeneous computing platforms. J. Biomed. Opt. 23 (1), 010504.
Pascual-Marqui, R.D., Michel, C.M., Lehmann, D., 1994. Low resolution electromagnetic Zhang, X., Toronov, V.Y., Webb, A.G., 2005. Simultaneous integrated diffuse optical
tomography: a new method for localizing electrical activity in the brain. Int. J. tomography and functional magnetic resonance imaging of the human brain. Opt.
Psychophysiol. https://doi.org/10.1016/0167-8760(84)90014-x. Express. 13 (14), 5513–5521.
Peirce, J., et al., 2019. PsychoPy2: experiments in behavior made easy. Behav. Res. Meth. Zhao, Y., et al., 2021. NIRS-ICA: a MATLAB toolbox for independent component analysis
51, 195–203. https://doi.org/10.3758/s13428-018-01193-y. applied in fNIRS studies. Front. Neuroinform. 15, 683735 https://doi.org/10.3389/
Penny, W.D., Friston, K.J., Ashburner, J.T., Kiebel, S.J., Nichols, T.E. (Eds.), 2006. fninf.2021.683735.
Statistical parametric mapping: the analysis of functional brain images. Academic Zijdenbos, A.P., et al., 1994. Morphometric analysis of white matter lesions in MR
Press. images: method and validation. IEEe Trans. Med. ImAging 13 (4), 716–724. https://
Saager, R., Berger, A., 2008. Measurement of layer-like hemodynamic trends in scalp and doi.org/10.1109/42.363096.
cortex: implications for physiological baseline suppression in functional near- Zou, K.H., et al., 2004. Statistical validation of image segmentation quality based on a
infrared spectroscopy. J. Biomed. Opt. 13 (3), 034017 https://doi.org/10.1117/ spatial overlap index. Acad. Radiol. 11 (2), 178–189. https://doi.org/10.1016/
1.2940587. s1076-6332(03)00671-8.

11

You might also like