diagnostics

Article
Dual-Energy CT, Virtual Non-Calcium Bone Marrow Imaging
of the Spine: An AI-Assisted, Volumetric Evaluation of a
Reference Cohort with 500 CT Scans
Philipp Fervers 1, *, Florian Fervers 2 , Mathilda Weisthoff 1 , Miriam Rinneburger 1 , David Zopfs 1 ,
Robert Peter Reimer 1 , Gregor Pahn 3 , Jonathan Kottlors 1 , David Maintz 1 , Simon Lennartz 1 , Thorsten Persigehl 1
and Nils Große Hokamp 1

1 Department of Diagnostic and Interventional Radiology, Faculty of Medicine and University Hospital
Cologne, University Cologne, 50937 Cologne, Germany; [email protected] (M.W.);
[email protected] (M.R.); [email protected] (D.Z.); [email protected] (R.P.R.);
[email protected] (J.K.); [email protected] (D.M.); [email protected] (S.L.);
[email protected] (T.P.); [email protected] (N.G.H.)
2 Fraunhofer Institute of Optronics, System Technologies and Image Exploitation (IOSB),
76131 Karlsruhe, Germany; [email protected]
3 Philips CT Clinical Science, 22335 Hamburg, Germany; [email protected]
* Correspondence: [email protected]

Abstract: Virtual non-calcium (VNCa) images from dual-energy computed tomography (DECT) have
shown high potential to diagnose bone marrow disease of the spine, which is frequently disguised



by dense trabecular bone on conventional CT. In this study, we aimed to define reference values


for VNCa bone marrow images of the spine in a large-scale cohort of healthy individuals. DECT
Citation: Fervers, P.; Fervers, F.;
was performed after resection of a malignant skin tumor without evidence of metastatic disease.
Weisthoff, M.; Rinneburger, M.;
Image analysis was fully automated and did not require specific user interaction. The thoracolumbar
Zopfs, D.; Reimer, R.P.; Pahn, G.;
Kottlors, J.; Maintz, D.; Lennartz, S.;
spine was segmented by a pretrained convolutional neuronal network. Volumetric VNCa data of
et al. Dual-Energy CT, Virtual the spine’s bone marrow space were processed using the maximum, medium, and low calcium
Non-Calcium Bone Marrow Imaging suppression indices. Histograms of VNCa attenuation were created for each exam and suppression
of the Spine: An AI-Assisted, setting. We included 500 exams of 168 individuals (88 female, patient age 61.0 ± 15.9). A total of
Volumetric Evaluation of a Reference 8298 vertebrae were segmented. The attenuation histograms’ overlap of two consecutive exams, as
Cohort with 500 CT Scans. a measure for intraindividual consistency, yielded a median of 0.93 (IQR: 0.88–0.96). As our main
Diagnostics 2022, 12, 671. https:// result, we provide the age- and sex-specific bone marrow attenuation profiles of a large-scale cohort
doi.org/10.3390/diagnostics12030671 of individuals with healthy trabecular bone structure as a reference for future studies. We conclude
Academic Editor: Nico Buls that artificial-intelligence-supported, fully automated volumetric assessment is an intraindividually
robust method to image the spine’s bone marrow using VNCa data from DECT.
Received: 16 February 2022
Accepted: 8 March 2022
Keywords: bone marrow; spine; tomography; X-ray computed; artificial intelligence
Published: 9 March 2022

Publisher’s Note: MDPI stays neutral

with regard to jurisdictional claims in
published maps and institutional affil- 1. Introduction
iations.
Computed tomography (CT) of the chest and abdomen is the recommended and most
frequently conducted imaging procedure for the staging of malignant disease [1–5]. Due to
rapid scanning times, good patient acceptance, and few contraindications, it is also among
Copyright: © 2022 by the authors.
the most commonly performed imaging procedures in the Western world, with ever-rising
Licensee MDPI, Basel, Switzerland. numbers [6–8]. A well-known limitation of contrast-enhanced CT, compared with, for
This article is an open access article instance, magnetic resonance imaging (MRI), positron emission tomography CT (PET/CT),
distributed under the terms and or bone scintigraphy, is its limited capacity to accurately diagnose metastatic disease of
conditions of the Creative Commons the spine. The dense trabecular structure of vertebral bodies impedes the assessment of
Attribution (CC BY) license (https:// underlying, malignant tumors without dominant osteolytic or osteoblastic components.
creativecommons.org/licenses/by/ In a recent meta-analysis, conventional CT yielded a pooled sensitivity and specificity of
4.0/). 0.77 and 0.83 for only the detection of spine metastasis [9]. Since the thoracolumbar spine

Diagnostics 2022, 12, 671. https://doi.org/10.3390/diagnostics12030671 https://www.mdpi.com/journal/diagnostics

Diagnostics 2022, 12, 671 2 of 12

is among the overall most common locations of bone metastases, this diagnostic gap of
conventional CT represents a major clinical limitation [10,11]. Additional MRI, PET/CT,
or scintigraphy significantly improves the detection of metastatic spine disease but at an
additional cost of radiation exposure or extended contraindications. Moreover, additional
exams delay diagnosis, increase patient discomfort, and raise economic expenses [9].
By now, dual-energy CT (DECT) is a widely available technology, which holds the
potential to narrow the diagnostic uncertainty of conventional CT concerning metastatic
spine disease [12–14]. DECT exploits the physical phenomenon that the interaction of an
X-ray beam with any absorbing material depends on a material-specific combination of the
photoelectric effect and Compton scattering. Approximating the contribution of either sep-
arately allows for material decomposition, e.g., for calcium [15,16]. In the context of spine
imaging, the postprocessing of virtual non-calcium (VNCa) images is of particular interest.
Using voxel-by-voxel material decomposition, VNCa images emulate Hounsfield units
(HUs) without the calcium-specific portion of X-ray attenuation [12]. This technique aims
to virtually remove the trabecular structure of the vertebral body and enables dedicated
bone marrow imaging of the spine [12]. Several recent studies suggest similar capabilities
of VNCa images, compared with the gold standards MRI and PET/CT, when assessing
malignant infiltration of the spine’s bone marrow [12,17–23] or traumatic vertebral bone
marrow alterations [24,25]. While study-specific cutoffs have been provided for these
purposes, knowledge on the distribution and variability of VNCa values encountered in
healthy individuals is sparse.
This lack of reference values for the physiological bone marrow attenuation on VNCa
images obligates each comparative study to define an individual reference cohort and
furthermore impairs real-world applicability of findings from the literature. Hence, the
aim of our study is to provide a large-scale reference cohort of physiological bone marrow
attenuation on VNCa, as a foundation to support further investigation of pathological bone
marrow alterations.

2. Materials and Methods

All procedures performed in studies involving human participants were conducted
in accordance with the ethical standards of the institutional (Application Number 21-
1105) and national research committee and with the 1964 Helsinki declaration and its
later amendments or comparable ethical standards. Informed consent was waived due to
retrospective study characteristics.

2.1. Patient Enrollment

Inclusion criteria to our study comprised the following factors:
(1) Intravenous contrast-enhanced DECT of the chest and abdomen by the protocol
specified below;
(2) Examination performed between 1 January 2016 and 1 January 2021 after resection of
a malignant skin tumor;
(3) Patient age >18 years.
Exclusion criterium was macroscopic evidence of metastatic disease.
Metastatic disease was excluded in consensus reading by two experienced radiologists
(with 4 and 7 years of experience in oncologic imaging). A total of 500 DECT scans
conforming to the enrollment criteria were randomly selected as our study population.

2.2. DECT Imaging Protocol

Patients were examined in a head-first, supine position on a commercially available
DECT scanner (IQon Spectral CT, Philips Healthcare, Amsterdam, The Netherlands). All
scans were performed after intravenous contrast administration in the portal venous phase,
using bolus tracking in the descending thoracic aorta (delay of 50 s, threshold 150 HU). Scan
parameters were as follows: tube voltage 120 kV; tube current modulated by DoseRight
3D-DOM (Philips Healthcare); collimation 64 × 0.625 mm; pitch 0.671; a total of 100 mL
Diagnostics 2022, 12, 671 3 of 12

of iodinated contrast medium (Accupaque, GE Healthcare) was administered by a 20 g

intravenous catheter with a flow rate of 3.0 mL/s, followed by a saline flush of 30 mL.

2.3. DECT Image Reconstruction and Postprocessing

All images were constructed in a 512 × 512 matrix and a slice thickness of 2 mm
with an overlap of 1 mm. Spectral-based raw data were processed to VNCa images using
the vendor’s proprietary software (IntelliSpace Portal, Spectral Diagnostics Suite, Philips
Healthcare) employing maximum, medium, and low calcium suppression indices (index
25, 50, and 76, respectively).

2.4. Automated Segmentation of the Bone Marrow

Data assessment in our study was fully automated and did not require specific user
interaction. First, the spine was segmented on conventional images employing a pretrained,
convolutional neuronal network, which won the VerSe Vertebrae Segmentation Challenge
by the Technical University of Munich in 2020 [26,27]. The neuronal network by Payer et al.
ranked first, with a dice coefficient of 0.94 and a correct vertebrae labeling rate of 0.99,
compared with expert readings of >300 CT examinations [28]. In the second step, excess ver-
tebrae above a maximum number of 17, counted from the bottom one (5 lumbar and 12 tho-
racic vertebrae), were identified by a Python script. We discarded excess vertebrae above
17 to avoid inconsistent inclusion of the partially imaged cervical spine. Consecutively,
the resulting volume of interest (VOI) containing the thoracolumbar spine was narrowed
at all margins by 3 mm, using the SciPy command “scipy.ndimage.binary_erosion”. This
process aimed to exclude the vertebrae’s bordering cortical bone, which does not contain
bone marrow, resulting in the three-dimensionally segmented thoracolumbar spine’s bone
marrow space. Lastly, the VOI mask was automatically transferred to the postprocessed
VNCa images at low, medium, and maximum calcium suppression settings (Figure 1).
Bone marrow attenuation histograms with a bin size of 5 HU were yielded by a Python
script at each suppression level. Data visualization was achieved by 3D Slicer [29].

2.5. Statistical Assessment

Statistical analysis was performed in the R language for statistical computing (version
4.0.0, R Foundation, Vienna, Austria). For further analysis, bone marrow attenuation
histograms were normalized to a standard volume. Attenuation values below −1000 HU in
the bone marrow space were not respected, since they appear due to VNCa postprocessing
of densely calcified structures (e.g., cortical bone islands, segments of cortical bone), which
do not contain bone marrow [23]. Attenuation histograms were reported by patient age,
sex, and the level of calcium suppression (low, medium, maximum). The intraindividual
consistency of our automated method was assessed by the percentage overlap of attenuation
histograms at two consecutive examinations. Similarly, the age- and sex-specific bone
marrow attenuation profiles were compared by the overlap of attenuation histograms
(Figure 2). Figures were plotted using the R library ggplot2 [30].
After segmentation of the thoracolumbar spine by a convolutional neuronal network,
intraindividual consistency (panel A) was assessed by the overlap of normalized bone mar-
row attenuation histograms in consecutive exams. Patient A underwent two dual-energy
CT scans within 119 days (gray and turquoise area), yielding an overlap of bone marrow
attenuation of 0.95, which locates close to the median overlap in our study population.
Panel B illustrates the interindividual overlap of bone marrow attenuation between males
and females in the age group of 21–40 years, presented by the sex- and age-specific average
attenuation histograms (overlap 0.81). It is worth noting that for reasons of simplicity, only
data with maximum calcium suppression levels are illustrated.
Diagnostics 2022, 12, 671 4 of 12

Figure 1. Volumetric assessment of bone marrow attenuation. Data assessment in our study was
achieved by a pretrained convolutional neuronal network and did not require specific user interaction.
Axial computed tomography slices served as input to the neuronal network by Payer et al. (A). After
automated segmentation, excess vertebrae above 17, counting from the most bottom one, were
excluded (B). Consecutively, the established volume of interest (VOI) was narrowed at each margin
by 3 mm, which aimed to partially exclude the bordering cortical bone but spare the bone marrow
space (C). Lastly, the VOI was transferred to the virtual non-calcium postprocessed data at three
different calcium suppression levels ((D), only maximum calcium suppression shown), resulting in
volumetric bone marrow attenuation data (E). Histograms of bone marrow attenuation were extracted
for further processing.
Diagnostics 2022, 12, 671 5 of 12

Figure 2. Intra- (A) and interindividual (B) overlap of automated, volumetric bone marrow attenua-
tion assessment.

3. Results
We evaluated 500 DECT scans of 168 patients (88 female, mean age 61.0 ± 15.9 years).
The median number of exams per patient was 2 [2–4], the median time interval between
two examinations was 185 [160–231] days. A total of 8298 thoracolumbar vertebrae were
segmented with a median of 17 [16,17] vertebrae per exam. This corresponds to the inclu-
sion of 97.6% of all possible thoracolumbar vertebrae (8298 segmented vertebrae/500 DECT
scans × 17 thoracolumbar vertebrae). Batch processing of DECT data was not interrupted
for manual alterations.
The intraindividual consistency of our method was assessed by the overlap of at-
tenuation histograms. The median overlap of bone marrow attenuation of consecutive
exams yielded an overall 92.9% [87.8–95.6]. The median intraindividual overlaps in low,
medium, and maximum VNCa settings were 93.7% [90.1–96.2], 91.6% [87.1–95.4], and 92.4%
[86.8–95.1], respectively. Notably, 92.8% and 65.3% of follow-up examinations achieved at
least a “good” and “excellent” intraindividual consistency, respectively (overlap of >80.0%
and >90.0%, respectively). The study sample and basic results are illustrated in Figure 3.

3.1. Quantitative Features of Physiological Bone Marrow Attenuation

Bone marrow attenuation was assessed on low, medium, as well as maximum calcium
suppression settings, and grouped by patient age and sex. Table 1 reports median attenua-
tion values along with quartiles and 95th percentiles, and the location of the histogram’s
maximum for each group. Comprehensive attenuation histograms are plotted in Figure 4.
Diagnostics 2022, 12, 671 6 of 12

Figure 3. Summary statistics and patient characteristics. Histogram of patient age (A). Mean age
of included patients was 61.0 ± 15.9 years. Dashed lines represent the borders of age subgroups
for further analysis. Pie chart of patient sex (B). Patient sex was balanced in our dataset, including
88 females vs. 80 males. Pie chart of automated segmentation results (C). We achieved segmentation
of 97.6% (n = 8298) of all included thoracolumbar vertebrae in our sample (n = 8500). Density chart
of attenuation histogram overlap (D). Intraindividual consistency of our method was assessed by
the overlap of attenuation histograms in consecutive exams. Median overlap was 92.9% [87.8–95.6].
92.9% of follow-up examinations yielded at least a “good” and 65.3% an “excellent” intraindividual
consistency (overlap > 80.0% and >90%, respectively).

Table 1. Quantitative features of physiological bone marrow attenuation in virtual non-calcium

(VNCa) data.

Male Female
Patient
Age Median 95th Perc. Median 95th Perc.
IQR (HU) Max. (HU) IQR (HU) Max. (HU)
(HU) (HU) (HU) (HU)
21–40 years −2.5 −22.5–12.5 57.5 −7.5 −17.5 −32.5–2.5 37.5 −17.5
41–60 years −7.5 −27.5–12.5 57.5 −12.5 −17.5 −32.5–2.5 42.5 −17.5
Low VNCa
61–80 years −12.5 −32.5–7.5 47.5 −17.5 −22.5 −42.5–−7.5 27.5 −27.5
>80 years −22.5 −37.5–2.5 42.5 −22.5 −27.5 −42.5–−7.5 32.5 −27.5
Diagnostics 2022, 12, 671 7 of 12

Table 1. Cont.

Male Female
Patient
Age Median 95th Perc. Median 95th Perc.
IQR (HU) Max. (HU) IQR (HU) Max. (HU)
(HU) (HU) (HU) (HU)
21–40 years −122.5 −152.5–92.5 −47.5 −117.5 −147.5 −177.5–−117.5 −77.5 −142.5
Medium 41–60 years −107.5 −142.5–−77.5 −32.5 −102.5 −117.5 −147.5–−92.5 −52.5 −112.5
VNCa 61–80 years −107.5 −142.5–−82.5 −37.5 −102.5 −107.5 −132.5–−82.5 −47.5 −102.5
>80 years −107.5 −142.5–−77.5 −37.5 −97.5 −97.5 −122.5–−72.5 −32.5 −87.5
21–40 years −422.5 −507.5–−347.5 −237.5 −417.5 −477.5 −567.5–−407.5 −312.5 −442.5
Maximum 41–60 years −362.5 −462.5–−287.5 −182.5 −327.5 −382.5 −467.5–−317.5 −212.5 −357.5
VNCa 61–80 years −342.5 −452.5–−272.5 −177.5 −307.5 −312.5 −397.5–−252.5 −167.5 −282.5
>80 years −322.5 −432.5–−247.5 −152.5 −282.5 −267.5 −352.5–−202.5 −122.5 −227.5
The physiological bone marrow attenuation is reported by median values, quartiles, 95th percentiles, and
histogram maxima, grouped by patient sex, age, and calcium suppression settings (low, medium, and maximum).

Figure 4. Attenuation histograms of physiological bone marrow in virtual non-calcium. Attenuation

histograms are plotted for low, medium, and maximum calcium suppression (three columns), for
males and females (two rows), and four age groups (four colors). The bone marrow space of each
patient was normalized to a standard volume, to achieve an equal area under the curve throughout
Diagnostics 2022, 12, 671 8 of 12

attenuation histograms for increased comparability. Bone marrow attenuation below −1000 HU was
excluded from this analysis since it appears due to virtual calcium suppression of densely calcified
structures (e.g., cortical bone, cortical bone islands), which do not contain bone marrow.

3.2. Age- and Sex-Specific Attenuation of Bone Marrow

Bone marrow attenuation distribution was not uniform but varied depending on
patient age and sex. The largest spread of bone marrow attenuation profiles was observed
in the maximum calcium suppression index (Figure 4), which is the most frequently used
VNCa setting for bone marrow imaging in recent literature [12,23]. In higher calcium
suppression levels, bone marrow attenuation was found generally lower in younger age.
The age-adjusted, sex-related differences in bone marrow attenuation were less pronounced
(Figure 5).

Figure 5. Cohort triangle of the age- and sex-specific discrepancy of bone marrow attenuation
profiles. The changes in bone marrow attenuation profiles were assessed by the overlap of volume-
standardized, average, virtual non-calcium attenuation histograms between all possible age- and
sex-specific combinations. A high overlap of attenuation profiles is plotted by a darker shade of
red. The largest discrepancy (=smallest overlap) was observed between the youngest vs. the oldest
included female individuals (female, 21–40 years vs. female, >80 years, overlap 0.35). Inter-sex,
age-adjusted overlap of female vs. male bone marrow attenuation was relatively high (0.81–0.89). It
is worth noting that for reasons of simplicity, the bone marrow attenuation is compared at maximum
calcium suppression level since this setting demonstrated the largest spread of attenuation profiles
and has been described as most valuable for the diagnosis of bone marrow disease [12,23].

4. Discussion
DECT has recently been suggested to overcome the limited diagnostic accuracy of CT
in diagnosing metastatic disease of the spine. Particularly, VNCa images postprocessed
from DECT data have shown the potential to close the diagnostic gap and suggested similar
capabilities to MRI and PET/CT for detection of spinal metastasis [12]. To facilitate the
clinical transition of reports on VNCa performance in diagnosing occult malignant disease
of the spine by conventional CT, this study sought to provide reference values for VNCa
attenuation of the spine’s bone marrow among healthy individuals across different age
groups and sexes, which are missing in literature to date.
Diagnostics 2022, 12, 671 9 of 12

In order to avoid bias, we avoided specific user interactions when identifying the
reference values. Artificial-intelligence-based data processing allowed for a fully auto-
mated methodology, aiming at maximum generalizability of our results. Notably, the
intraindividual overlap of attenuation profiles, as a check for consistency of our methodol-
ogy, was excellent (median overlap of consecutive bone marrow attenuation histograms
92.9%). The excellent consistency in consecutive examinations suggests that VNCa bone
marrow attenuation is a robust, reproducible imaging parameter. As our main result, we
provide an in-depth description of the normal ranges of age- and sex-specific bone marrow
attenuation profiles.
A recent study has successfully identified an attenuation cutoff on VNCa images
for detection of metastatic disease of the spine but without acknowledgment of the age-
and sex-specific changes in physiological bone marrow attenuation [12]. Abdullayev et al.
investigated a cutoff between −174.9 HU and −143.2 HU on maximum VNCa suppression
settings for diagnosis of spinal metastasis, using contrast-enhanced DECT in a relatively
young patient population (mean age 56 years, 71.4% female) [12]. Using identical technol-
ogy and a similar imaging protocol, our reference cohort supports their findings for the
evaluated patient cohort, since both cutoffs locate above the respective age- and sex-specific
95th percentile of physiological bone marrow attenuation (age 41–60 years, female: refer-
ence 95th percentile −212.5 HU, Table 1). Malignant infiltration displaces the fatty, healthy
bone marrow by soft tissue tumor, raising the bone marrow’s VNCa attenuation [17].
This hypothesis has been recently investigated for bone marrow infiltration by multiple
myeloma on non-contrast-enhanced VNCa images [18,20,23]. A shift in VNCa attenuation
above the 95th percentile of the here-suggested age- and sex-matched reference range in the
context of malignant disease might serve as a desirable imaging biomarker in the future.
Definition of the upper margin of a normal reference range at the 95th percentile is
a common practice to achieve an arbitrary distinction between a physiological vs. patho-
logical result of a medical test [31–34]. Adjustment for basic patient demographics, such
as age and sex, however, is crucial for reliable results [33,34]. Recalling the study by
Abdullayev et al., the investigated cutoffs for the malignant disease were above the 95th
percentile of their study sample; however, they were located in the normal range of bone
marrow attenuation of elderly females (age > 80 years, female: reference 95th percentile
−122.5 HU, Table 1), rendering them physiological in such patients [12]. Our data suggest a
large disparity of bone marrow attenuation between younger and elderly individuals, par-
ticularly pronounced among women. Hence, without adjustment for sex and age, arbitrary
cutoffs for the diagnosis of bone marrow disease are of limited use.
Traditionally, bone marrow imaging was a domain of MRI, and several investigations
aimed to outline the normal ranges of physiological bone marrow appearance [35,36]. In
particular, MRI is feasible to image the physiological conversion from cellular, red bone
marrow to fatty, yellow marrow in healthy individuals [35]. Our data demonstrate higher
bone marrow attenuation with rising age, which cannot be explained by fatty conversion
since fat demonstrates lower attenuation than cell-rich tissues. Hence, we suggest that the
age-dependent increase in bone marrow attenuation in our study is a result of decreasing
bone mineral density, i.e., virtual calcium suppression of voxels with higher calcium-like
attenuation results in relatively lower VNCa attenuation. Similarly, iodine uptake of the
bone marrow might affect VNCa attenuation: Thus far, most VNCa studies of malignant
bone marrow disease are restricted to non-contrast-enhanced scans [17–23]. A possible
explanation is that virtual suppression of calcium-like attenuation might interfere with
iodine-like attenuation [12,37]. In particular, when using higher VNCa settings, besides the
desired suppression of calcium-like attenuation, a portion of the iodine-like attenuation
is also virtually suppressed [12]. This technical aspect of VNCa postprocessing, however,
is a well-known and reproducible limitation [12,38,39]. However, compared with our
investigation, which aimed to examine the overlap of material-specific attenuation, a
recent phantom study using an identical spectral detector CT (SDCT) scanner suggested
excellent capability to separate calcium- and iodine-like attenuation [40]. Abdullayev
Diagnostics 2022, 12, 671 10 of 12

et al. also made use of the identical SDCT scanner to successfully identify metastatic spine
disease on contrast-enhanced VNCa images, which might serve as well as a practical proof
of concept [12]. Several further authors consider this approach promising and warrant
further investigation [13,14].
Our study has several limitations that need to be discussed. First, the values of
normal ranges we report are limited to the specific scanner used; nevertheless, we consider
our methodology transferable to other imaging protocols. Second, while macroscopic
tumor burden was excluded based on image assessment, all patients had a history of
local dermatological malignancy (which was considered successfully treated). However,
microscopic tumor burden might be present at the timepoint of imaging. Last, the analyzed
patient population demonstrated a wide demographic spread; however, it was not truly
normally distributed; the age and sex distribution resembled patient populations commonly
encountered in oncologic imaging in daily practice.

5. Conclusions
In conclusion, we provided the first, large-scale reference cohort of healthy individuals
in VNCa bone marrow imaging from SDCT. We outlined age- and sex-specific normal
ranges of bone marrow attenuation to facilitate the clinical transfer of DECT-based assess-
ment of metastatic disease of the spine.

Author Contributions: Conceptualization, P.F. and N.G.H.; methodology, P.F., G.P. and F.F.; software,
F.F. and G.P.; validation, D.M., T.P. and N.G.H.; formal analysis, P.F.; investigation, M.R., M.W., J.K.,
R.P.R. and D.Z.; resources, T.P., N.G.H. and S.L.; data curation, P.F.; writing—original draft prepara-
tion, P.F.; writing—review and editing, N.G.H., R.P.R., D.Z., D.M., J.K., S.L. and T.P.; visualization,
P.F.; supervision, N.G.H., D.M. and S.L.; project administration, N.G.H.; funding acquisition, N.G.H.
All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: The study was conducted in accordance with the Declaration
of Helsinki and approved by the Institutional Review Board of the University of Cologne (protocol
code 21-1105 on 8 March 2021).
Informed Consent Statement: Patient consent was waived by the local review board due to retro-
spective study design.
Data Availability Statement: All data are reported comprehensively in the manuscript’s result section.
Conflicts of Interest: N.G.H. and D.M. are on the speaker’s bureau of Philips Healthcare. D.Z.
received institutional research support unrelated to this project from Philips Healthcare. G.P. is an
employee of Philips Healthcare. The funders had no role in the design of the study; in the collection,
analyses, or interpretation of data; in the writing of the manuscript; in the decision to publish
the results.

References
1. Schmoll, H.J.; Van Cutsem, E.; Stein, A.; Valentini, V.; Glimelius, B.; Haustermans, K.; Nordlinger, B.; van de Velde, C.J.; Balmaña, J.;
Regula, J.; et al. ESMO Consensus Guidelines for management of patients with colon and rectal cancer. A personalized approach
to clinical decision making. Ann. Oncol. 2012, 23, 2479–2516. [CrossRef]
2. Garbe, C.; Amaral, T.; Peris, K.; Hauschild, A.; Arenberger, P.; Bastholt, L.; Bataille, V.; Del Marmol, V.; Dréno, B.; Fargnoli, M.C.;
et al. European consensus-based interdisciplinary guideline for melanoma. Part 1: Diagnostics—Update 2019. Eur. J. Cancer 2020,
126, 141–158. [CrossRef] [PubMed]
3. Loutfi, S.; Khankan, A.; Al Ghanim, S. Guidelines for multimodality radiological staging of lung cancer. J. Infect. Public Health
2012, 5, S14–S21. [CrossRef] [PubMed]
4. Kwee, T.C.; Kwee, R.M.; Nievelstein, R.A.J. Imaging in staging of malignant lymphoma: A systematic review. Blood 2008, 111,
504–516. [CrossRef] [PubMed]
5. Kalli, S.; Semine, A.; Cohen, S.; Naber, S.P.; Makim, S.S.; Bahl, M. American Joint Committee on Cancer’s Staging System for
Breast Cancer, Eighth Edition: What the Radiologist Needs to Know. Radiographics 2018, 38, 1921–1933. [CrossRef] [PubMed]
6. Statista. Most Frequent Radiology Procedures U.S. 2018. Available online: https://www.statista.com/statistics/1030144/most-
frequent-radiology-procedures-in-the-us/ (accessed on 8 January 2022).
Diagnostics 2022, 12, 671 11 of 12

7. Statistics »Diagnostic Imaging Dataset. Available online: https://www.england.nhs.uk/statistics/statistical-work-areas/

diagnostic-imaging-dataset/ (accessed on 8 January 2022).
8. Smith-Bindman, R.; Miglioretti, D.L.; Larson, E.B. Rising Use of Diagnostic Medical Imaging in a Large Integrated Health System:
The use of imaging has skyrocketed in the past decade, but no one patient population or medical condition is responsible. Health
Aff. 2008, 27, 1491. [CrossRef] [PubMed]
9. Yang, H.-L.; Liu, T.; Wang, X.-M.; Xu, Y.; Deng, S.-M. Diagnosis of bone metastases: A meta-analysis comparing 18FDG PET, CT,
MRI and bone scintigraphy. Eur. Radiol. 2011, 21, 2604–2617. [CrossRef]
10. Soeharno, H.; Povegliano, L.; Choong, P.F. Multimodal Treatment of Bone Metastasis—A Surgical Perspective. Front. Endocrinol.
2018, 9, 518. [CrossRef]
11. Kakhki, V.R.D.; Anvari, K.; Sadeghi, R.; Mahmoudian, A.-S.; Kahkki, M.T. Pattern and distribution of bone metastases in common
malignant tumors. Nucl. Med. Rev. 2013, 16, 66–69. [CrossRef]
12. Abdullayev, N.; Hokamp, N.G.; Lennartz, S.; Holz, J.A.; Romman, Z.; Pahn, G.; Neuhaus, V.; Maintz, D.; Krug, B.; Borggrefe, J. Im-
provements of diagnostic accuracy and visualization of vertebral metastasis using multi-level virtual non-calcium reconstructions
from dual-layer spectral detector computed tomography. Eur. Radiol. 2019, 29, 5941–5949. [CrossRef]
13. D’Angelo, T.; Albrecht, M.H.; Caudo, D.; Mazziotti, S.; Vogl, T.J.; Wichmann, J.L.; Martin, S.; Yel, I.; Ascenti, G.; Koch, V.; et al.
Virtual non-calcium dual-energy CT: Clinical applications. Eur. Radiol. Exp. 2021, 5, 38. [CrossRef] [PubMed]
14. Kim, C.; Kim, W.; Park, S.-J.; Lee, Y.H.; Hwang, S.H.; Yong, H.S.; Oh, Y.-W.; Kang, E.-Y.; Lee, K.Y. Application of Dual-Energy
Spectral Computed Tomography to Thoracic Oncology Imaging. Korean J. Radiol. 2020, 21, 838–850. [CrossRef] [PubMed]
15. Patino, M.; Prochowski, A.; Agrawal, M.D.; Simeone, F.J.; Gupta, R.; Hahn, P.F.; Sahani, D.V. Material Separation Using
Dual-Energy CT: Current and Emerging Applications. Radiographics 2016, 36, 1087–1105. [CrossRef] [PubMed]
16. Hokamp, N.G.; Salem, J.; Hesse, A.; Holz, J.A.; Ritter, M.; Heidenreich, A.; Maintz, D.; Haneder, S. Low-Dose Characterization of
Kidney Stones Using Spectral Detector Computed Tomography: An Ex Vivo Study. Investig. Radiol. 2018, 53, 457–462. [CrossRef]
17. Kosmala, A.; Weng, A.M.; Krauss, B.; Knop, S.; Bley, T.A.; Petritsch, B. Dual-energy CT of the bone marrow in multiple myeloma:
Diagnostic accuracy for quantitative differentiation of infiltration patterns. Eur. Radiol. 2018, 28, 5083–5090. [CrossRef]
18. Kosmala, A.; Weng, A.; Heidemeier, A.; Krauss, B.; Knop, S.; Bley, T.A.; Petritsch, B. Multiple Myeloma and Dual-Energy CT:
Diagnostic Accuracy of Virtual Noncalcium Technique for Detection of Bone Marrow Infiltration of the Spine and Pelvis. Radiology
2018, 286, 205–213. [CrossRef]
19. Palmer, W.E.; Simeone, F.J. Can Dual-Energy CT Challenge MR Imaging in the Diagnosis of Focal Infiltrative Bone Marrow
Lesions? Radiology 2018, 286, 214–216. [CrossRef] [PubMed]
20. Thomas, C.; Schabel, C.; Krauss, B.; Weisel, K.; Bongers, M.; Claussen, C.D.; Horger, M. Dual-Energy CT: Virtual Calcium
Subtraction for Assessment of Bone Marrow Involvement of the Spine in Multiple Myeloma. Am. J. Roentgenol. 2015, 204,
W324–W331. [CrossRef]
21. Fervers, P.; Glauner, A.; Gertz, R.; Täger, P.; Kottlors, J.; Maintz, D.; Borggrefe, J. Virtual calcium-suppression in dual energy
computed tomography predicts metabolic activity of focal MM lesions as determined by fluorodeoxyglucose positron-emission-
tomography. Eur. J. Radiol. 2020, 135, 109502. [CrossRef]
22. Fervers, P.; Celik, E.; Bratke, G.; Maintz, D.; Baues, C.; Ruffing, S.; Pollman-Schweckhorst, P.; Kottlors, J.; Lennartz, S.;
Hokamp, N.G. Radiotherapy Response Assessment of Multiple Myeloma: A Dual-Energy CT Approach with Virtual Non-
Calcium Images. Front. Oncol. 2021, 11, 3766. [CrossRef]
23. Fervers, P.; Fervers, F.; Kottlors, J.; Lohneis, P.; Pollman-Schweckhorst, P.; Zaytoun, H.; Rinneburger, M.; Maintz, D.; Hokamp, N.G.
Feasibility of artificial intelligence–supported assessment of bone marrow infiltration using dual-energy computed tomography
in patients with evidence of monoclonal protein—A retrospective observational study. Eur. Radiol. 2021, 12, 1–11. [CrossRef]
24. Wang, C.-K.; Tsai, J.-M.; Chuang, M.-T.; Wang, M.-T.; Huang, K.-Y.; Lin, R.-M. Bone Marrow Edema in Vertebral Compression
Fractures: Detection with Dual-Energy CT. Radiology 2013, 269, 525–533. [CrossRef] [PubMed]
25. Gu, L.; Liang, J.; Zhan, Y.; Pan, J.; Zhang, Y.; Cai, W.; Hong, G. Dual-Energy Computed Tomography Virtual Noncalcium Imaging
for the Detection of Acute Bone Marrow Edema in Vertebrae: Qualitative and Quantitative Analysis. J. Med. Imaging Health
Inform. 2021, 11, 752–759. [CrossRef]
26. Sekuboyina, A.; Bayat, A.; Husseini, M.E.; Löffler, M.; Li, H.; Tetteh, G.; Kukačka, J.; Payer, C.; Štern, D.; Urschler, M.; et al.
VerSe: A Vertebrae Labelling and Segmentation Benchmark for Multi-Detector CT Images; Elsevier: Amsterdam, The Netherlands, 2020;
Available online: https://opus4.kobv.de/opus4-zib/frontdoor/index/index/docId/8090 (accessed on 12 January 2021).
27. Payer, C.; Štern, D.; Bischof, H.; Urschler, M. Coarse to Fine Vertebrae Localization and Segmentation with SpatialConfiguration-
Net and U-Net. In Proceedings of the 15th International Joint Conference on Computer Vision, Imaging and Computer Graphics
Theory and Applications, Valletta, Malta, 27–29 February 2020; SCITEPRESS—Science and Technology Publications: Setúbal,
Portugal, 2020; pp. 124–133. [CrossRef]
28. Leaderboard—Grand Challenge. Available online: https://verse2020.grand-challenge.org/evaluation/challenge/leaderboard/
(accessed on 11 January 2022).
29. Kikinis, R.; Pieper, S.D.; Vosburgh, K.G. 3D Slicer: A Platform for Subject-Specific Image Analysis, Visualization, and Clinical
Support. In Intraoperative Imaging and Image-Guided Therapy; Springer: New York, NY, USA, 2014; pp. 277–289. [CrossRef]
30. Wickham, H. ggplot2: Elegant Graphics for Data Analysis; Springer: New York, NY, USA, 2016; ISBN 978-3-319-24277-4. Available
online: https://ggplot2.tidyverse.org (accessed on 7 January 2022).
Diagnostics 2022, 12, 671 12 of 12

31. Taveras, E.M.; Rifas-Shiman, S.L.; Sherry, B.; Oken, E.; Haines, J.; Kleinman, K.; Rich-Edwards, J.W.; Gillman, M.W. Crossing
Growth Percentiles in Infancy and Risk of Obesity in Childhood. Arch. Pediatr. Adolesc. Med. 2011, 165, 993–998. [CrossRef]
32. Keller, K.; Sinning, C.; Schulz, A.; Jünger, C.; Schmitt, V.H.; Hahad, O.; Zeller, T.; Beutel, M.; Pfeiffer, N.; Strauch, K.; et al. Right
atrium size in the general population. Sci. Rep. 2021, 11, 22523. [CrossRef] [PubMed]
33. Harrington, D.M.; Staiano, A.; Broyles, S.T.; Gupta, A.K.; Katzmarzyk, P.T. BMI percentiles for the identification of abdominal
obesity and metabolic risk in children and adolescents: Evidence in support of the CDC 95th percentile. Eur. J. Clin. Nutr. 2012,
67, 218–222. [CrossRef] [PubMed]
34. Daniels, S.R. How to Define Hypertension in Children and Adolescents. Circulation 2016, 133, 350–351. [CrossRef] [PubMed]
35. Colombo, A.; Bombelli, L.; Summers, P.E.; Saia, G.; Zugni, F.; Marvaso, G.; Grimm, R.; Jereczek-Fossa, B.A.; Padhani, A.R.;
Petralia, G. Effects of sex and age on fat fraction, diffusion-weighted image signal intensity and apparent diffusion coefficient in
the bone marrow of asymptomatic individuals: A cross-sectional whole-body MRI study. Diagnostics 2021, 11, 913. [CrossRef]
36. Liney, G.P.; Bernard, C.P.; Manton, D.J.; Turnbull, L.W.; Langton, C.M. Age, gender, and skeletal variation in bone marrow
composition: A preliminary study at 3.0 Tesla. J. Magn. Reson. Imaging 2007, 26, 787–793. [CrossRef]
37. Zopfs, D.; Lennartz, S.; Zaeske, C.; Merkt, M.; Laukamp, K.R.; Reimer, R.P.; Maintz, D.; Borggrefe, J.; Hokamp, N.G. Phantomless
assessment of volumetric bone mineral density using virtual non-contrast images from spectral detector computed tomography.
Br. J. Radiol. 2020, 93, 20190992. [CrossRef]
38. Borggrefe, J.; Neuhaus, V.-F.; Le Blanc, M.; Hokamp, N.G.; Maus, V.; Mpotsaris, A.; Lennartz, S.; Dos Santos, D.P.; Maintz, D.;
Abdullayev, N. Accuracy of iodine density thresholds for the separation of vertebral bone metastases from healthy-appearing
trabecular bone in spectral detector computed tomography. Eur. Radiol. 2018, 29, 3253–3261. [CrossRef] [PubMed]
39. Sauter, A.P.; Muenzel, D.; Dangelmaier, J.; Braren, R.; Pfeiffer, F.; Rummeny, E.J.; Noël, P.; Fingerle, A.A. Dual-layer spectral
computed tomography: Virtual non-contrast in comparison to true non-contrast images. Eur. J. Radiol. 2018, 104, 108–114.
[CrossRef] [PubMed]
40. Hua, C.-H.; Shapira, N.; Merchant, T.E.; Klahr, P.; Yagil, Y. Accuracy of electron density, effective atomic number, and iodine
concentration determination with a dual-layer dual-energy computed tomography system. Med. Phys. 2018, 45, 2486–2497.
Available online: https://onlinelibrary.wiley.com/doi/full/10.1002/mp.12903 (accessed on 7 January 2022). [CrossRef] [PubMed]