Modeling the Properties of White Matter Tracts Using Diffusion Tensor Imaging to Characterize Patterns of Injury in Aging and Neurodegenerative Disease

Abstract

Diffusion tensor imaging (DTI) is a relatively novel magnetic resonance-based imaging methodology that can provide valuable insight into the microstructure of white matter tracts of the brain. In this paper, we evaluated the reliability and reproducibility of deriving a semi-automated pseudo-atlas DTI tractography method vs. standard atlas-based analysis alternatives, for use in clinical cohorts with neurodegeneration and ventriculomegaly. We showed that the semi-automated pseudo-atlas DTI tractography method was reliable and reproducible across different cohorts, generating 97.7% of all tracts. However, DTI metrics obtained from both methods were significantly different across the majority of cohorts and white matter tracts ( p < 0.001). Despite this, we showed that both methods produced patterns of white matter injury that are consistent with findings reported in the literature and with DTI profiles generated from these methodologies. Scatter plots comparing DTI metrics obtained from each methodology showed that the pseudo-atlas method produced metrics that implied a more preserved neural structure compared to its counterpart. When comparing DTI metrics against a measure of ventriculomegaly (i.e., Evans’ Index), we showed that the standard atlas-based method was able to detect decreasing white matter integrity with increasing ventriculomegaly, while in contrast, metrics obtained using the pseudo-atlas method were sensitive for stretch or compression in the posterior limb of the internal capsule. Additionally, both methods were able to show an increase in white matter disruption with increasing ventriculomegaly, with the pseudo-atlas method showing less variability and more specificity to changes in white matter tracts near to the ventricles. In this study, we found that there was no true gold-standard for DTI methodologies or atlases. Whilst there was no congruence between absolute values from DTI metrics, differing DTI methodologies were still valid but must be appreciated to be variably sensitive to different changes within white matter injury occurring concurrently. By combining both atlas and pseudo-atlas based methodologies with DTI profiles, it was possible to navigate past such challenges to describe white matter injury changes in the context of confounders, such as neurodegenerative disease and ventricular enlargement, with transparency and consistency.

Full text

fnagi-14-787516 April 21, 2022 Time: 14:29 # 1
ORIGINAL RESEARCH
published: 27 April 2022
doi: 10.3389/fnagi.2022.787516
Edited by:
Ping Wu,
Fudan University, China
Reviewed by:
Eric Schmidt,
Université de Toulouse, France
Ian Brian Malone,
University College London,
United Kingdom
*Correspondence:
Nicole C. Keong
nchkeong@cantab.net
Specialty section:
This article was submitted to
Neurocognitive Aging and Behavior,
a section of the journal
Frontiers in Aging Neuroscience
Received: 30 September 2021
Accepted: 15 March 2022
Published: 27 April 2022
Citation:
Kok CY, Lock C, Ang TY and
Keong NC (2022) Modeling
the Properties of White Matter Tracts
Using Diffusion Tensor Imaging
to Characterize Patterns of Injury
in Aging and Neurodegenerative
Disease.
Front. Aging Neurosci. 14:787516.
doi: 10.3389/fnagi.2022.787516
Modeling the Properties of White
Matter Tracts Using Diffusion Tensor
Imaging to Characterize Patterns of
Injury in Aging and
Neurodegenerative Disease
Chun Yen Kok1, Christine Lock2, Ting Yao Ang2and Nicole C. Keong1,2*
1Duke-National University of Singapore (NUS) Medical School, Singapore, Singapore, 2Department of Neurosurgery,
National Neuroscience Institute, Singapore, Singapore
Diffusion tensor imaging (DTI) is a relatively novel magnetic resonance-based imaging
methodology that can provide valuable insight into the microstructure of white matter
tracts of the brain. In this paper, we evaluated the reliability and reproducibility of
deriving a semi-automated pseudo-atlas DTI tractography method vs. standard atlas-
based analysis alternatives, for use in clinical cohorts with neurodegeneration and
ventriculomegaly. We showed that the semi-automated pseudo-atlas DTI tractography
method was reliable and reproducible across different cohorts, generating 97.7% of all
tracts. However, DTI metrics obtained from both methods were significantly different
across the majority of cohorts and white matter tracts (p<0.001). Despite this, we
showed that both methods produced patterns of white matter injury that are consistent
with findings reported in the literature and with DTI profiles generated from these
methodologies. Scatter plots comparing DTI metrics obtained from each methodology
showed that the pseudo-atlas method produced metrics that implied a more preserved
neural structure compared to its counterpart. When comparing DTI metrics against a
measure of ventriculomegaly (i.e., Evans’ Index), we showed that the standard atlas-
based method was able to detect decreasing white matter integrity with increasing
ventriculomegaly, while in contrast, metrics obtained using the pseudo-atlas method
were sensitive for stretch or compression in the posterior limb of the internal capsule.
Additionally, both methods were able to show an increase in white matter disruption
with increasing ventriculomegaly, with the pseudo-atlas method showing less variability
and more specificity to changes in white matter tracts near to the ventricles. In this
study, we found that there was no true gold-standard for DTI methodologies or atlases.
Whilst there was no congruence between absolute values from DTI metrics, differing
DTI methodologies were still valid but must be appreciated to be variably sensitive
to different changes within white matter injury occurring concurrently. By combining
Frontiers in Aging Neuroscience | www.frontiersin.org 1April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 2
Kok et al. Modeling DTI Injury Patterns
both atlas and pseudo-atlas based methodologies with DTI profiles, it was possible
to navigate past such challenges to describe white matter injury changes in the context
of confounders, such as neurodegenerative disease and ventricular enlargement, with
transparency and consistency.
Keywords: diffusion tensor imaging (DTI), white matter, region of interest (ROI), tractography,Alzheimer’s disease,
ventriculomegaly
INTRODUCTION
Diffusion tensor imaging (DTI) is a relatively novel magnetic
resonance-based imaging methodology that maps the water
diffusion properties within the brain (Mori and Zhang, 2006).
Since water generally diffuses along intact white matter tracts
of the brain, the diffusion properties can therefore provide
information about the microarchitecture of specific white matter
tracts in the brain. DTI metrics that can be obtained consist
of fractional anisotropy (FA), mean diffusivity (MD), axial
diffusivity (L1) and radial diffusivity (L2 and 3).
DTI has been used to investigate patterns of white matter
changes at a microstructural level in various cohorts, such
as normal pressure hydrocephalus (NPH), optic nerve
decompression and in the developing human brain (Lebel
et al., 2008;Paul et al., 2014;Keong et al., 2017). Diffusion
tensor metrics have been shown to be reliable biomarkers for
Alzheimer’s disease progression (Acosta-Cabronero et al., 2012),
and are also sensitive to changes in white matter injury and
compression in patients with NPH after surgical intervention
(Scheel et al., 2012;Keong et al., 2017).
However, DTI acquisition, processing, and analysis is a
complex multi-step process that is subject to many variables
which may affect the results and interpretation thereof
(Mukherjee et al., 2008;Soares et al., 2013;Christidi et al.,
2016). The post-processing and analysis of DTI metrics is
non-trivial and dependent on the availability of software and
infrastructure. Quantitative DTI metrics can be obtained
by various methods including tract-based spatial statistics
(TBSS) which is a voxel-based morphometry-like approach;
or the manual placement of 2D region of interest (ROI);
atlas-registration based parcellations using a pre-defined
white matter atlas to describe tracts of interest (Smith et al.,
2006;Mukherjee et al., 2008;Oishi et al., 2009;Soares et al.,
2013). or per-image automated tractography approaches, such
as TRACULA (Yendiki et al., 2011). ROI analyses are time
consuming, influenced by inter-rater variability, and subject to
variations along a tract. The Alzheimer’s Disease Neuroimaging
Initiative (ADNI) group have previously published (Nir et al.,
2013) on the use of both (i) white matter tract atlas ROIs,
i.e., registration of images from a DTI atlas to each subject’s
distortion corrected FA image, before applying an atlas of white
matter labels and superimposing these atlas ROIs into the same
coordinate space as subject results for analysis and (ii) TBSS tract
atlas ROIs as per (Smith et al., 2006). Per-image automated DTI
tractography approaches are an attractive method for disease
specific cohorts but are dependent on accurate registration
and may be confounded by anatomical differences attributed
to neurodegenerative diseases like Alzheimer’s disease and the
distortions of white matter tracts secondary to the presence of
significant ventriculomegaly, such as in NPH (Mukherjee et al.,
2008;Zalesky, 2011;Scheel et al., 2012;Acosta-Cabronero and
Nestor, 2014).
In this paper, we evaluated the reliability and reproducibility
of differing automated DTI tractography methods to produce
diffusion metrics of various white matter tracts in the presence of
known confounders such as atrophy in aging, neurodegeneration
and significant ventriculomegaly. We firstly aimed to develop a
cohort-specific pseudo atlas-based semi-automated tractography
method that was comparable to an atlas-based DTI analysis
currently utilized by the ADNI group; as we interrogated ADNI
datasets for this study, we have therefore defined the latter
method as the “gold-standard” approach for reference. We found
that the diffusion metrics generated from the former were
significantly different from those generated by the latter.
We hypothesized that the results from differing DTI
methodologies could be subject to the impact of different
algorithmic modifications. To test our hypotheses, we designed
the following experiments to optimize the application of DTI
methodologies to describe white matter injury patterns in the
presence of confounders such as neurodegenerative disease and
degree of ventriculomegaly:
1. We developed a cohort-specific pseudo atlas-based semi-
automated tractography to generate white matter tracts of
interest and compared it to that of the “gold standard”
atlas-based DTI analysis currently utilized by ADNI, in
order to assess the reproducibility and reliability of the
novel methodology.
2. We performed initial comparisons on this pilot to examine
the agreement of DTI metrics obtained from white matter
tracts generated by both methodologies.
a. Due to the poor agreement of the metrics, we proceeded
to test the differing DTI methodologies under different
processing algorithms to assess how these impacted the
agreement of the metrics generated.
3. We performed testing using a known model of white
matter at-risk of injury. This ROI model allowed us
to test for white matter distortion patterns in three
cohorts of patients with different levels of confounders,
namely varying degrees of neurodegeneration and
atrophy along the spectrum from cognitively normal to
Alzheimer’s disease.
4. In addition, we performed testing to examine the effect of
increasing ventriculomegaly on this ROI model of white
matter at-risk. We performed independent quantification
Frontiers in Aging Neuroscience | www.frontiersin.org 2April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 3
Kok et al. Modeling DTI Injury Patterns
of ventricular size by measuring the Evans’ index (EI)
and concurrently examined changes in DTI metrics in the
context of increasing ventriculomegaly for the ROI model
of white matter tracts at risk.
5. Finally, we performed a further layer of analysis to
confirm the diffusivity changes in this study by generating
morphological patterns of DTI metrics for independent
interpretation. We utilized DTI profiles, radar graphs
of all DTI metrics (FA, MD, L1, L2, and 3), in order
to illustrate differences between DTI methodologies and
across cohorts. The concept and utility of DTI profiles
has been previously described by our group in describing
patterns of white matter injury across clinical cohorts
(Lock et al., 2018).
MATERIALS AND METHODS
Data used in the preparation of this article were obtained
from the ADNI database.1ADNI DTI metrics used for
comparison were the UCLA DTI ROI summary measures for
ADNIGO and ADNI2.
Subjects
The ADNI study recruited patients between the ages of 55 and 90
from 57 sites in the United States and Canada. For this study, we
retrospectively selected subjects who had screening/baseline MRI
scans with diffusion-weighted images (DWI) from the ADNI
image data archive. The selected scans included 51 cognitively
normal (CN) subjects (mean age 72.47 ±6.13 years; 45.1%
male), 48 patients with Alzheimer’s disease (AD) (75.00 ±8.67
years; 58.3% male) and 70 patients with early mild cognitive
impairment (EMCI) (72.71 ±8.29 years; 61.4% male). These
cohorts were selected because we wanted to interrogate the
methodologies in patients with varying degrees of cognitive
impairment and atrophy.
Open-Source Software Used
3D slicer is an open source quantitative imaging network tool we
used to derive ventricular morphological indices and to conduct
3D volumetric segmentation (Fedorov et al., 2012). MRIcroGL
is an open source software developed by Neuroimaging Tools
and Resources Collaboratory (NITRC) used in this study to
convert DICOM images to NIfTI format [(Nitrc) N.T.R.C, 2014].
ExploreDTI is a graphical toolbox written in MATLAB that
was used in this project for DTI and white matter tractography
(Leemans et al., 2009).
MRI Acquisition and Post-processing
MRI scans were performed on 3T GE Medical Systems scanners
across participating ADNI sites. Diffusion scans were acquired
with 256 ×256 matrix; voxel size 2.7 mm ×2.7 mm ×2.7 mm; 41
DWI (b= 1,000 s/mm2) and 5 b0 images. More information on
the MRI protocol is available at http://adni.loni.usc.edu/methods/
documents/mri-protocols/.
1adni.loni.usc.edu
Pre-processing was required to convert each patient’s set
of unsorted DICOM format axial DWI images where two-
dimensional DICOM image slices were converted into a single 3D
NIfTI file with MRIcroGL. During this step, 1 subject in the CN
cohort was excluded due to a failure to convert it from DICOM
to NIfTI format.
DTI files were generated using ExploreDTI. Thereafter,
they were corrected for subject motion and eddy current
induced geometric distortion. Whole brain tractography
was then performed.
White Matter Tracts
Utilizing the known ROI model of white matter at-risk, we chose
to analyze 8 unique white matter tracts. Bilaterally, we analyzed a
total of 14 white matter tracts, and they were as follows: Body of
the corpus callosum (bCC), Genu of the corpus callosum (gCC),
Inferior fronto-occipital Fasciculus (IFO), Inferior Longitudinal
Fasciculus (ILF), Anterior Thalamic Radiation (ATR), Posterior
Thalamic Radiation (PTR), Posterior Limb of the Internal
Capsule (PLIC), and Uncinate Fasciculus (UF) (Hofer and
Frahm, 2006;Wakana et al., 2007;Oishi et al., 2010;Borden
et al., 2015;Keong et al., 2017). The bCC and gCC are midline
structures while the rest are found bilaterally. Therefore, with
48 DWI in the AD cohort and 50 DWI in the CN cohort, this
amounted to a total of 672 and 700 white matter tracts in the AD
and CN cohort, respectively.
Methods of Automated Diffusion Tensor
Imaging Tractography
In this paper, two methods of automated tractography were
compared. The first method was a cohort-specific pseudo atlas-
based semi-automated tractography method (termed Method
1) where a randomly selected image in each cohort is used as
a template for white matter tract generation in the remaining
images. The second method was an automated atlas-based ROI
analysis (termed Method 2) where a standardized lab-based atlas
was used as the template. These two methods were tested on
the AD and CN cohorts. We followed this up by implementing
additional algorithmic modifications to assess if they affected the
results of the methodologies.
We implemented two modifications to the processing
algorithms. The first was to try an alternative standardized
atlas as a template in Method 2 (using the alternative atlas
template is termed Method 3). The second was to optimize
the alignment to the ACPC plane prior to performing the
DTI analysis by following protocol adapted from the Human
Connectome Project (HCP) pre-processing pipelines (Glasser
et al., 2013). This was done by co-registration of the DWI
to the MNI template (standardized template from 152 subject
scans) and the corresponding T1-weighted image. This allowed
all images in the dataset to be oriented and aligned to the
same space such that the anterior and posterior commissures
(ACPC) were aligned along a horizontal plane. By ensuring that
all images in the dataset were standardized in terms of position
and orientation, we sought to improve the fit of both atlases
(and the pseudo-atlas) as applied to the images in the dataset.
Frontiers in Aging Neuroscience | www.frontiersin.org 3April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 4
Kok et al. Modeling DTI Injury Patterns
Once we compared and found the technical considerations to
have improved the tractography, we subsequently applied the
refined methodology to all available cohorts to complete our DTI
analysis, with the exception of the EMCI cohort, where only the
ACPC alignment was enacted. This was because earlier results
from AD and CN cohorts already showed that ACPC alignment
improved the tract analysis success rate but did not fully eliminate
the large differences between methodologies, so we proceeded to
refine our analysis by only using ACPC aligned EMCI scans.
Method 1: Automated Atlas Based Tractography
We randomly chose a representative subject in each cohort and
set its FA map as a “pseudo-atlas.” To ensure that it was suitable
as a template, the image was subject to visual inspection as
a quality check and compared to other images to ensure that
there were no obvious defects and distortions. Using ExploreDTI,
specific white matter tracts of the pseudo-atlas were generated
from user-determined regions of interests (ROIs). The type of
ROI placed enforced different conditions within the area enclosed
by the ROI. Placing an AND ROI generated tracts that passed
through this area. Placing a NOT ROI excluded fibers that passed
through this area. Table 1 Shows the types of ROIs and their
respective locations which were used to isolate the corresponding
white matter tracts.
With the pseudo-atlas and ROIs as input, the software applied
similar ROIs to the remaining images in the cohort using a
deterministic streamline method (Lebel et al., 2008). White
matter tracts were then automatically reconstructed via the
automatically generated ROIs in the remaining images of the
cohort. Where the ROIs extruded to an image failed to generate
any tracts, this resulted in missing data.
Methods 2 and 3: Automated Atlas-Based Region of
Interest Analysis
A widely used standardized DTI template with its associated
white matter tracts was used as an atlas. The white matter tracts in
the atlas were generated from ROIs determined by the template
creator. The atlas template was warped, and the associated ROIs
transformed and applied to each image in the dataset. The
diffusion metrics were then automatically generated from the
resulting white matter tracts defined by the ROIs. The problem
of missing data as in Method 1 was also found to occur using this
method but was not as significant as in Method 1.
Method 2 utilized the ICBM-DTI-81 atlas from the ICBM DTI
workgroup (Oishi et al., 2008). This atlas template was created by
averaging hand segmentation of diffusion tensor maps from 81
subjects with a mean age of 39 with 42 males and 39 females.
Method 3 utilized the JHU white matter tractography atlas
from the Laboratory of Brain Anatomical MRI, Johns Hopkins
University (Oishi et al., 2009). This atlas was based on averaging
results from running deterministic tractography on 28 normal
subjects with a mean age of 29 with 17 males and 11 females.
The atlas used in Method 2 contained all 8 unique white
matter tracts we wanted to investigate whereas the atlas used in
TABLE 1 | ROI constraints used to isolate white matter tracts in the pseudo-atlas for Method 1.
Tract ROI constraints
gCC - Sagittal AND: Define anterior 1/6 of the length of the corpus callosum.
- Parasagittal NOT: Slices lateral to corticospinal tract bilaterally, defining entire slice.
bCC - Splenium of corpus callosum consists of the posterior 1/4 of the length of corpus callosum.
- Sagittal AND: Define remaining length of the corpus callosum excluding the genu and splenium—from 1/6 to 3/4 length of corpus callosum.
- Axial NOT: Slice just beneath the bCC, defining entire slice.
ATR - Coronal AND: Slice chosen in the middle of the gCC, defining anterior limb of internal capsule.
- Coronal AND: Slice at the anterior edge of pons, defining entire thalamus.
- Sagittal NOT: Defining entire central slice.
- Coronal NOT: Slice at the posterior thalamic edge, defining entire slice.
IFO - Coronal AND: Slice at the anterior edge of gCC, defining entire slice.
- Coronal AND: Slice at the halfway mark of parieto-occipital sulcus, defining the occipital lobe.
- Sagittal NOT: Define entire central slice.
ILF - Coronal AND: Slice at the posterior edge of cingulum, defining occipital lobe.
- Coronal AND: Most posterior coronal slice in which the temporal lobe is not connected to the frontal lobe (as seen on axial view), defining the anterior
temporal lobe.
- Coronal NOT: Same slice as above, defining the rest of the brain except anterior temporal lobe.
- Sagittal NOT: Defining entire central slice.
PLIC - Axial AND: Slice where PLIC is visibly the largest, defining the PLIC.
- Axial AND: Slice at the inferior slice where the PLIC is still visible, defining the PLIC.
- Axial NOT: Slice at the condensed portion of the corticospinal tract in the brain stem, defining entire slice.
PTR - Coronal AND: Slice at the posterior edge of the cingulum, defining anterior-posterior directing, periventricular white matter tracts.
- Parasagittal AND: Slice at the lateral edge of thalamus, defining entire thalamus.
- Coronal NOT: Slice at the anterior edge of thalamus, defining entire slice.
- Axial NOT: slice at the inferior edge of thalamus, defining entire slice.
UF - Axial AND: Slice where condensed cephalic-caudal directed fibers are distinct in the temporal lobe, defining temporal lobe.
- Coronal AND: Slice anterior to the condensed cephalic-caudal directed fibers, defining inferior frontal lobe.
- Coronal AND: Same slice as above, defining temporal lobe
- Coronal NOT: Slice posterior to the condensed cephalic-caudal directed fibers, defining entire slice.
gCC, genu of the corpus callosum; bCC, body of the corpus callosum; ATR, anterior thalamic radiation; IFO, inferior fronto-occipital fasciculus; ILF, inferior longitudinal
fasciculus; PLIC, posterior limb of the internal capsule; PTR, posterior thalamic radiation; UF, uncinate fasciculus.
Frontiers in Aging Neuroscience | www.frontiersin.org 4April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 5
Kok et al. Modeling DTI Injury Patterns
Method 3 only identified 5 of the 8 tracts we required: gCC, ATR,
IFO, ILF, and UF.
Characterizing
Ventriculomegaly—Evans’ Index
The Evans’ index (EI) is commonly used to characterize the
degree of ventriculomegaly from a patient’s MRI or CT scan
(Yamada et al., 2016). It is defined as the ratio of the maximum
width of the frontal horns of the lateral ventricles to the
maximum internal width of the cranial vault as seen on the axial
view (Yamada et al., 2016). 3D Slicer was used to re-align T1 scans
to the ACPC for consistency and derive the EI (Soon et al., 2021).
Statistical Analysis
Diffusion metrics FA and MD from both left and right tracts
were averaged for the purposes of the analysis. Paired t-tests were
used to test for differences between the different methodologies.
Linear correlation was used to investigate the association
between diffusion metrics and ventriculomegaly measured by the
Evans’ index. Scatter plots of FA and MD obtained from both
methodologies for all tracts across the AD, EMCI, and CN cohorts
were plotted to show the agreement of metrics and the variance
within each methodology. All statistical analyses were performed
using R statistical software (version 4.0.4) (R Development
Core Team, 2010). A p-value of <0.05 was considered to be
statistically significant.
Diffusion Tensor Imaging Profiles
DTI profiles are presented as radar graphs of means of all DTI
metrics (FA, MD, L1, L2, and 3), in order to provide a simplistic
illustration of differences between the various methods, as well
as differences across the spectrum of disease for AD. We have
previously demonstrated the utility of DTI profiles to describe
and compare disease processes in white matter tracts across
different cohorts (Keong et al., 2017;Lock et al., 2018). White
matter tract profiles were also generated using Tract Analysis
Profiles to illustrate how DTI metrics vary along each white
matter tract (Yeatman et al., 2012).
RESULTS
Reproducibility and Reliability of
Methodologies
Method 1 generated 96.3% (647/672) of all white matter tracts in
the AD cohort and 96.9% (678/700) in the CN cohort. Method 1
was unable to generate 25 tracts in the AD cohort and 22 tracts
in the CN cohort. This is in contrast to Method 2 which was able
to generate 99.6% (669/672) and 100% (700/700) tracts in the AD
and CN cohorts, respectively. This amounts to 3 missing tracts in
the AD cohort. After the implementation of the ACPC alignment,
the reliability and reproducibility of Method 1 improved with
98.7% (663/672) and 98.9% (692/700) success rate in the AD and
CN cohort, respectively. There were 9 missing tracts in the former
and 8 in the latter. Method 2 generated 100% of tracts in both
AD and CN cohorts. Implementation of Methods 1 and 2 on the
ACPC aligned-EMCI cohort likewise showed high success rates
of 96.2% (916/952) and 99.9% (951/952), respectively.
Comparison of Diffusion Tensor Imaging
Metrics Across Methodologies
Tables 2,3show the results of the paired t-tests conducted on
the DTI metrics obtained from the two methodologies across
all 8 white matter tracts. Table 2 compares the FA and MD
obtained using Method 1 with those using Methods 2 and 3
applied on non-ACPC aligned images in the AD and CN cohorts,
respectively. Table 3 also compares Methods 1 with 2 and 3 but
applied on scans that have undergone the ACPC alignment and
include scans from the EMCI cohort.
Non-ACPC aligned Method 1 was not well correlated to
Methods 2 and 3 (Table 2). FA and MD from Method 1 were
significantly different (p<0.001) from Methods 2 and 3 for all
tracts in AD and CN cohorts, except for PTR MD in CN. After
ACPC alignment and co-registration, Method 1 was significantly
different (p<0.001) from Methods 2 and 3 for all tracts in
AD, CN, and EMCI cohorts, except for PTR MD in AD and
CN (Table 3).
Figure 1 compares the FA of ACPC aligned images and non-
ACPC aligned images obtained using Methods 2 and 3 against
Method 1 in the 8 white matter tracts across CN and AD cohorts.
Figure 2 compares the MD of ACPC aligned images and non-
ACPC aligned images obtained using Methods 2 and 3 against
Method 1. Figure 3 compares both the FA and MD of ACPC
aligned images using Methods 2 and 3 against Method 1 in the
EMCI cohort. Non-ACPC aligned images in the EMCI cohort
were not compared here because the results were similar to AD
and CN images. The findings here seem to generally agree with
those from the paired t-tests.
FA and MD scatter plots demonstrated poor agreement
between Method 1 vs. 2 and Method 1 vs. 3 across all tracts in
AD and CN (Figures 1,2). This was not improved even with
ACPC alignment and co-registration of images. These trends are
also present with the addition of a cohort with an intermediate
severity of disease process (i.e., EMCI).
Across the paired t-tests in Tables 2,3and scatter plots in
Figures 1–3, FA and MD obtained using Methods 1, 2, and 3
show poor agreement and consistency across the CN, AD, and
EMCI cohorts. This is evidenced by the low linear correlation
coefficients and relatively large mean differences in the metrics
obtained across all tracts and cohorts as well as the scatterplots
showing a large deviation from the 45-degree diagonal line.
This demonstrates that the type of standard atlases used in
Method 2 (i.e., an alternative atlas was also tested using Method
3) did not meaningfully improve the agreement. Additionally,
comparing across the AD and CN cohorts also showed no
changes in agreements. Implementing the ACPC alignment
across all scans improved the agreements across all tracts
only marginally. Notably, the inter-methodological differences
were greater than the differences due to the application of
technical considerations. This confirmed the fact that there were
external confounding factors impacting the methodologies which
rendered them incomparable.
Frontiers in Aging Neuroscience | www.frontiersin.org 5April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 6
Kok et al. Modeling DTI Injury Patterns
White Matter Pattern Changes in
Cognitively Normal vs. Early Mild
Cognitive Impairment vs. Alzheimer’s
Disease Cohorts
Despite the lack of agreement, Methods 1, 2, and 3 showed
general trends that illustrate and reaffirm the presence of
different white matter pattern changes when comparing across
the AD, EMCI, and CN cohorts with varying degrees of
neurodegeneration. Figures 1–3show that there was greater
variability for white matter structures adjacent or near to the
ventricles such as the bCC and gCC as well as multidirectional
tracts like the PLIC and UF. The difference, however, is
that Method 1 reports white matter tracts having generally
higher FA and lower MD values, implying a more preserved
neural structure, compared to the other two methods for
each cohort tested.
Effect on White Matter Pattern Changes
With Increasing Ventriculomegaly
Figure 4 shows the scatter plots of FA and MD obtained from
both Methods 1 and 2 plotted against the Evans’ index (EI)
for all 8 white matter tracts across all 3 cohorts of AD, EMCI,
and CN combined. Only the ACPC aligned images are used in
this analysis due to its superior reliability and reproducibility
as previously shown. From the figures, we observed differing
patterns of correlation with EI when using Method 1 compared
to Method 2. From metrics obtained using Method 2, as EI
increased (implying increasing ventriculomegaly) there was a
TABLE 2 | Comparison of FA and MD derived by Method 1 against Methods 2 and 3 (non-ACPC aligned and co-registered) across white matter tracts in the (A)
Alzheimer’s disease cohort and (B) cognitively normal cohort.
(A) AD cohort
Linear correlation Significance of correlation Mean difference Paired t-test significance
Tract 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3
bCC FA 0.150 0.308 –0.099 <0.001
MD 0.514 <0.001 0.0004 <0.001
gCC FA 0.226 0.494 0.127 <0.001 –0.144 –0.237 <0.001 <0.001
MD 0.511 0.557 <0.001 <0.001 0.0003 0.0004 <0.001 <0.001
ATR FA 0.416 0.378 0.004 0.009 –0.136 –0.136 <0.001 <0.001
MD 0.502 0.281 <0.001 0.056 0.0005 0.0005 <0.001 <0.001
IFO FA 0.469 0.474 <0.001 <0.001 –0.213 –0.208 <0.001 <0.001
MD 0.421 0.406 0.003 0.005 0.0002 0.0002 <0.001 <0.001
ILF FA 0.291 0.091 0.047 0.542 –0.236 –0.230 <0.001 <0.001
MD 0.124 0.138 0.408 0.355 0.0002 0.0002 <0.001 <0.001
PLIC FA 0.145 0.330 –0.086 <0.001
MD 0.054 0.716 0.0002 <0.001
PTR FA 0.352 0.015 –0.100 <0.001
MD 0.298 0.042 0.0001 <0.001
UF FA –0.045 0.237 0.762 0.109 –0.202 –0.207 <0.001 <0.001
MD 0.263 0.396 0.074 0.006 0.0008 0.0005 <0.001 <0.001
(B) CN cohort
Linear correlation Significance of correlation Mean difference Paired t-test significance
Tract 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3
bCC FA 0.415 0.003 –0.066 <0.001
MD 0.502 <0.001 0.0003 <0.001
gCC FA 0.354 0.602 0.012 <0.001 –0.124 –0.234 <0.001 <0.001
MD 0.520 0.459 <0.001 <0.001 0.0002 0.0003 <0.001 <0.001
ATR FA 0.340 0.301 0.017 0.036 –0.127 –0.128 <0.001 <0.001
MD 0.481 0.545 <0.001 <0.001 0.0004 0.0004 <0.001 <0.001
IFO FA 0.547 0.505 <0.001 <0.001 –0.217 –0.213 <0.001 <0.001
MD 0.607 0.554 <0.001 <0.001 0.0001 0.0001 <0.001 <0.001
ILF FA 0.427 0.290 0.002 0.041 –0.235 –0.227 <0.001 <0.001
MD 0.571 0.532 <0.001 <0.001 0.0001 0.0001 <0.001 <0.001
PLIC FA –0.022 0.879 –0.058 <0.001
MD 0.182 0.205 0.0001 <0.001
PTR FA 0.174 0.232 –0.086 <0.001
MD 0.479 <0.001 0.0000 0.296
UF FA –0.086 0.445 0.554 0.001 –0.205 –0.206 <0.001 <0.001
MD 0.154 0.389 0.286 0.005 0.0005 0.0004 <0.001 <0.001
A negative mean difference indicates that FA/MD derived by Method 1 is higher than that of Methods 2 or 3. All mean difference of MD is in mm2/s.
Frontiers in Aging Neuroscience | www.frontiersin.org 6April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 7
Kok et al. Modeling DTI Injury Patterns
TABLE 3 | Comparison of FA and MD derived by Method 1 against Methods 2 and 3 (ACPC aligned and co-registered) across white matter tracts in the (A) Alzheimer’s
disease cohort, (B) cognitively normal cohort, and (C) early mild cognitive impairment cohort.
(A) AD cohort
Linear correlation Significance of correlation Mean difference Paired t-test significance
Tract 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3
bCC FA 0.271 0.062 –0.073 <0.001
MD 0.506 <0.001 0.0004 <0.001
gCC FA 0.204 0.509 0.163 <0.001 –0.127 –0.244 <0.001 <0.001
MD 0.686 0.561 <0.001 <0.001 0.0003 0.0004 <0.001 <0.001
ATR FA 0.325 0.422 0.024 0.003 –0.131 –0.133 <0.001 <0.001
MD 0.715 0.775 <0.001 <0.001 0.0005 0.0005 <0.001 <0.001
IFO FA 0.550 0.537 <0.001 <0.001 –0.212 –0.215 <0.001 <0.001
MD 0.307 0.323 0.034 0.025 0.0002 0.0002 <0.001 <0.001
ILF FA 0.122 0.173 0.408 0.241 –0.235 –0.233 <0.001 <0.001
MD 0.302 0.311 0.037 0.032 0.0002 0.0002 <0.001 <0.001
PLIC FA 0.104 0.484 –0.074 <0.001
MD 0.312 0.031 0.0002 <0.001
PTR FA 0.337 0.019 –0.086 <0.001
MD 0.406 0.004 0.0000 0.060
UF FA 0.115 0.446 0.436 0.001 –0.202 –0.210 <0.001 <0.001
MD 0.281 0.392 0.053 0.006 0.0009 0.0005 <0.001 <0.001
(B) CN cohort
Linear correlation Significance of correlation Mean difference Paired t-test significance
Tract 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3
bCC FA 0.302 0.033 –0.055 <0.001
MD 0.503 <0.001 0.0004 <0.001
gCC FA 0.268 0.426 0.062 0.002 –0.109 –0.245 <0.001 <0.001
MD 0.556 0.510 <0.001 <0.001 0.0002 0.0003 <0.001 <0.001
ATR FA 0.488 0.473 <0.001 <0.001 –0.116 –0.120 <0.001 <0.001
MD 0.688 0.672 <0.001 <0.001 0.0004 0.0004 <0.001 <0.001
IFO FA 0.548 0.424 <0.001 0.002 –0.213 –0.215 <0.001 <0.001
MD 0.730 0.683 <0.001 <0.001 0.0001 0.0002 <0.001 <0.001
ILF FA 0.331 0.317 0.019 0.025 –0.234 –0.233 <0.001 <0.001
MD 0.583 0.618 <0.001 <0.001 0.0001 0.0001 <0.001 <0.001
PLIC FA –0.191 0.184 –0.047 <0.001
MD 0.181 0.208 0.0001 <0.001
PTR FA 0.114 0.429 –0.069 <0.001
MD 0.437 0.002 0.0000 0.449
UF FA –0.088 0.479 0.545 <0.001 –0.199 –0.209 <0.001 <0.001
MD 0.123 0.422 0.394 0.002 0.0005 0.0004 <0.001 <0.001
(C) EMCI cohort
Linear correlation Significance of correlation Mean difference Paired t-test significance
Tract 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3 1 vs. 2 1 vs. 3
bCC FA 0.334 0.006 –0.048 <0.001
MD 0.436 <0.001 0.0004 <0.001
gCC FA 0.414 0.553 <0.001 <0.001 –0.106 –0.245 <0.001 <0.001
MD 0.420 0.440 <0.001 <0.001 0.0002 0.0003 <0.001 <0.001
ATR FA 0.460 0.483 <0.001 <0.001 –0.127 –0.129 <0.001 <0.001
MD 0.227 0.535 0.069 <0.001 0.0005 0.0005 <0.001 <0.001
IFO FA 0.628 0.597 <0.001 <0.001 –0.208 –0.209 <0.001 <0.001
MD 0.572 0.523 <0.001 <0.001 0.0002 0.0002 <0.001 <0.001
ILF FA 0.502 0.558 <0.001 <0.001 –0.231 –0.230 <0.001 <0.001
MD 0.531 0.564 <0.001 <0.001 0.0001 0.0001 <0.001 <0.001
PLIC FA 0.114 0.361 –0.275 <0.001
MD 0.360 0.003 0.0001 <0.001
PTR FA 0.438 <0.001 –0.063 <0.001
MD 0.570 <0.001 0.0000 0.008
UF FA 0.161 0.562 0.197 <0.001 –0.193 –0.205 <0.001 <0.001
MD 0.398 0.580 0.001 <0.001 0.0006 0.0004 <0.001 <0.001
A negative mean difference indicates that FA/MD derived by Method 1 is higher than that of Methods 2 or 3. All mean difference of MD is in mm2/s.
Frontiers in Aging Neuroscience | www.frontiersin.org 7April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 8
Kok et al. Modeling DTI Injury Patterns
FIGURE 1 | Distribution of scatter plots of FA obtained using Methods 2 and 3 against Method 1, comparing non-ACPC aligned and co-registered with ACPC
aligned white matter tracts across cognitively normal (CN) and Alzheimer’s Disease (AD) cohorts. The ellipses assume a multivariate normal distribution with the
mean at the center and area of the ellipse representing 95% confidence level. gCC, genu of the corpus callosum; bCC, body of the corpus callosum; ATR, anterior
thalamic radiation; IFO, inferior fronto-occipital fasciculus; ILF, inferior longitudinal fasciculus; PLIC, posterior limb of the internal capsule; PTR, posterior thalamic
radiation; UF, uncinate fasciculus. (A) Comparing ACPC and Non-ACPC: FA from Method 1 vs. 2 in CN cohort. (B) Comparing ACPC and Non-ACPC: FA from
Method 1 vs. 3 in CN cohort. (C) Comparing ACPC and Non-ACPC: FA from Method 1 vs. 2 in AD cohort. (D) Comparing ACPC and Non-ACPC: FA from Method 1
vs. 3 in AD cohort.
significant decrease in FA which is consistent with decreased
white matter integrity. By contrast, metrics obtained using
Method 1, showed that with increasing EI, there was a significant
increase in FA seen in the PLIC which is consistent with patterns
of stretch or compression. Both methods showed significant
increases in MD with increasing EI, suggesting an increase in
global, multi-directional white matter disruption. Method 1,
however, showed less variability and was more specific to changes
in white matter tracts near to the ventricles.
Correlation to Diffusion Tensor Imaging
Profiles of White Matter Tracts
DTI profiles for the gCC and the UF tracts were selected
to illustrate differences between the methods and between
cohorts. Figure 5 shows that the difference in DTI metrics
generated by Methods 1, 2, and 3 can be distinguished with
DTI profiles. Method 1 consistently produced the lowest MD,
L1 and L2, and 3 values, compared to Methods 2 and 3.
Profiles for DTI metrics before and after ACPC alignment
were nearly visually indistinguishable for Methods 1 and 2.
Failure to correct for ACPC alignment did not influence the
results as much as the variation produced by the different
methods. DTI profiles demonstrated cohort differences between
AD, CN, and EMCI, across the spectrum of disease, but
inter-methodological differences were larger than inter-cohort
differences (Figure 6A). Likewise, change in DTI morphology
in the AD cohort after 12 months was not as pronounced as
inter-methodological differences (Figure 6B). White matter tract
profiles in Figure 7 show the variability of DTI metrics along
different white matter tracts.
DISCUSSION
In this paper, we demonstrated that it was possible to reliably
develop and refine an SOP for a pseudo atlas-based semi-
automated tractography DTI analysis method in the presence of
confounders comprising aging, neurodegenerative disease, and
ventricular enlargement. However, the absolute values of the DTI
metrics generated by this novel methodology did not align well
with those generated by standardized atlas-based DTI analyses,
despite implementing a differential of algorithmic modifications.
Regardless, we managed to show that the inter-methodological
differences between DTI metrics obtained from Method 1 and
2 were greater than the effects of implementing the algorithmic
Frontiers in Aging Neuroscience | www.frontiersin.org 8April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 9
Kok et al. Modeling DTI Injury Patterns
FIGURE 2 | Distribution of scatter plots of MD obtained using Methods 2 and 3 against Method 1, comparing non-ACPC aligned and co-registered with ACPC
aligned white matter tracts across cognitively normal (CN) and Alzheimer’s Disease (AD) cohorts. The ellipses assume a multivariate normal distribution with the
mean at the center and area of the ellipse representing 95% confidence level. gCC, genu of the corpus callosum; bCC, body of the corpus callosum; ATR, anterior
thalamic radiation; IFO, inferior fronto-occipital fasciculus; ILF, inferior longitudinal fasciculus; PLIC, posterior limb of the internal capsule; PTR, posterior thalamic
radiation; UF, uncinate fasciculus. (A) Comparing ACPC and Non-ACPC: MD from Method 1 vs. 2 in CN cohort. (B) Comparing ACPC and Non-ACPC: MD from
Method 1 vs. 3 in CN cohort. (C) Comparing ACPC and Non-ACPC: MD from Method 1 vs. 2 in AD cohort. (D) Comparing ACPC and Non-ACPC: MD from Method
1 vs. 3 in AD cohort.
modifications. Whilst this suggests that DTI output metrics
from differing methodologies cannot be directly compared for
statistical analysis, we also showed that DTI methodologies
were differentially impacted upon by confounders affecting
structural brain or ventricular changes. In the presence of such
considerations, we found that there was no true “gold-standard”
but rather, the differing methodologies were sensitive to differing
significant findings on a spectrum from contiguous to non-
contiguous changes, in ways that were both complementary to
each other and consistent with differences between such cohorts
as reported in published literature. Nevertheless, by creating the
DTI profiles from metrics generated by the methodologies, we
showed that, despite differing DTI values, the morphology of DTI
changes was consistent across DTI analysis methods.
A Cohort-Specific Pseudo Atlas-Based
Semi-Automated Tractography Method
vs. Standardized Atlas-Based Diffusion
Tensor Imaging Analysis
In our study, we found that a novel pseudo atlas-based semi-
automated tractography DTI analysis method (Method 1) was
reliable and reproducible. This was evidenced by the high success
rate of generating white matter tracts across both AD and CN
sub cohorts. Upon implementing the ACPC alignment (one
modification to the algorithm), the number of missing tracts
decreased from 25 to 9 tracts in the AD cohort and 22–8 tracts
in the CN cohorts. This showed that the intracohort variability
in image orientation could be a main contributor to the missing
tracts and that this refinement improved the reliability and
reproducibility of the methodology.
Surprisingly, we could not show that the actual DTI
metrics generated from the pseudo atlas-based semi-automated
tractography DTI analysis method (Method 1) were exactly
comparable to the standardized atlas-based DTI analysis
(Methods 2). This was despite implementing the both
modifications to the algorithm including an alternative published
and verified atlas (Method 3) and applying the ACPC alignment.
From observing the scatter plots in Figures 1–3we noted that
this disagreement can be attributed to Method 1 reporting
white matter tracts as having generally higher FA and lower
MD values compared to those obtained via Methods 2 and 3
across AD, CN, and EMCI cohorts. However, as this finding was
consistent across the varying spectrums of disease and aging,
Frontiers in Aging Neuroscience | www.frontiersin.org 9April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 10
Kok et al. Modeling DTI Injury Patterns
FIGURE 3 | Distribution of scatter plots of FA and MD obtained using Methods 2 and 3 against Method 1 (ACPC aligned and co-registered) across white matter
tracts in the early mild cognitive impairment (EMCI) cohort. The ellipses assume a multivariate normal distribution with the mean at the center and area of the ellipse
representing 95% confidence level. gCC, genu of the corpus callosum; bCC, body of the corpus callosum; ATR, anterior thalamic radiation; IFO, inferior
fronto-occipital fasciculus; ILF, inferior longitudinal fasciculus; PLIC, posterior limb of the internal capsule; PTR, posterior thalamic radiation; UF, uncinate fasciculus.
(A) Comparing ACPC and Non-ACPC: FA from Method 1 vs. 2 in EMCI cohort. (B) Comparing ACPC and Non-ACPC: FA from Method 1 vs. 3 in EMCI cohort. (C)
Comparing ACPC and Non-ACPC: MD from Method 1 vs. 2 in EMCI cohort. (D) Comparing ACPC and Non-ACPC: MD from Method 1 vs. 3 in EMCI cohort.
i.e., in AD (neurodegenerative), EMCI (mild neurodegenerative)
and CN (aging) cohorts, it suggests that a cohort-specific white
matter template (the pseudo-atlas) as employed in Method 1
was more sensitive to generating white matter tracts in the
presence of confounders compared to Methods 2 and 3. This
showed that while Method 1 was internally consistent across
cohorts, DTI output measures may not be directly comparable
to DTI measures from Methods 2 and 3 for purposes of
statistical analysis.
The Effect of Ventriculomegaly on the
Degree of Patterns of White Matter
Change
From Figure 4, we noted that with increasing EI, signifying an
increasing degree of ventriculomegaly, Method 1 showed that
PLIC had a significant increase in FA, Method 2 showed a
significant decrease in FA across all tracts and both Methods
1 and 2 showed significant increases in MD across all tracts.
An increase in FA is consistent with tracts under stretch or
compression while a decrease in FA is consistent with decreased
white matter integrity. Conversely, a rise in MD across all tracts
suggested an increase in global, multidirectional white matter
disruption. These patterns of white matter injury in the setting
of ventriculomegaly are consistent with findings reported in the
literature (Keong et al., 2017). Whilst the FA changes in PLIC
reflected in Methods 1 and 2 appear to be contradictory, such
conflict is consistent with DTI findings of previous work in
hydrocephalus where it was shown that FA can increase and
decrease within the same context, depending on the reversibility
of white matter injury (Assaf et al., 2006;Hattori et al., 2011;
Kanno et al., 2011;Ben-Sira et al., 2015;Keong et al., 2017;Tan
et al., 2018). This represents an important fallacy of interpreting
DTI changes based solely on global measures, such as FA or
MD alone. In particular, FA is highly dependent upon relative
changes in diffusivity measures; it can be driven to higher or
lower values based on predominant changes in axial diffusivity
over radial diffusivity and vice versa. In this study, we found
that the different patterns that were reflected in both methods
could be interpreted as complementary to each other. For
example, Method 2 may have detected the reduced white matter
integrity and hence decreased FA, whereas Method 1 detected
the compressive mechanism of injury and hence increased FA.
Method 1, however, showed less variability and was more specific
to changes in white matter tracts nearer to the ventricles (i.e.,
bCC, gCC), when compared to Method 2. These known DTI
Frontiers in Aging Neuroscience | www.frontiersin.org 10 April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 11
Kok et al. Modeling DTI Injury Patterns
FIGURE 4 | Scatter plots of FA and MD against Evans’ index (EI) across white matter tracts for CN, AD, and EMCI cohorts combined. gCC, genu of the corpus
callosum; bCC, body of the corpus callosum; ATR, anterior thalamic radiation; IFO, inferior fronto-occipital fasciculus; ILF, inferior longitudinal fasciculus; PLIC,
posterior limb of the internal capsule; PTR, posterior thalamic radiation; UF, uncinate fasciculus. (A) FA vs. Evans’ Index: bCC. (B) FA vs. Evans’ Index: gCC. (C) FA
vs. Evans’ Index: ATR. (D) FA vs. Evans’ Index: IFO. (E) FA vs. Evans’ Index: ILF. (F) FA vs. Evans’ Index: PLIC. (G) FA vs. Evans’ Index: PTR. (H) FA vs. Evans’
Index: UF. (I) MD vs. Evans’ Index: bCC. (J) MD vs. Evans’ Index: gCC. (K) MD vs. Evans’ Index: ATR. (L) MD vs. Evans’ Index: IFO. (M) MD vs. Evans’ Index: ILF.
(N) MD vs. Evans’ Index: PLIC. (O) MD vs. Evans’ Index: PTR. (P) MD vs. Evans’ Index: UF.
conflicts impacting upon the transparency and consistency of
interpretation of DTI results across literature would benefit from
the application of a more standardized common taxonomy; this
is an approach we have proposed elsewhere (Keong et al., 2017,
manuscript in submission).
Correlation to Diffusion Tensor Imaging
Profiles
The DTI profiles in Figures 5,6align with the above findings.
We showed that Method 1 reports more preserved white
matter profiles in comparison to Methods 2 and 3 across
AD, CN and EMCI cohorts. This supports the suggestion
that a cohort-specific template (the pseudo-atlas) was more
sensitive to demonstrating white matter integrity in the presence
of confounders due to aging and neurodegenerative disease.
Additionally, we found evidence that ACPC alignment did not
significantly affect the morphology of DTI profiles generated and
that inter-methodological differences were indeed larger than
inter-cohort differences. These cohorts include the spectrum
of Alzheimer’s disease, from CN to EMCI and finally to AD.
Inter-methodological differences were similar to or greater than
changes in DTI profiles in the AD cohort after 12 months.
Despite the variability of DTI values along the tracts (Figure 7)
as well as between methodologies (as seen in DTI profiles),
the morphology of the DTI profile still remains consistent
across cohorts and aligns well with published literature. This
lack of comparability in DTI analysis methodologies and
variability, ultimately supports the use of DTI profiles in the
analysis of DTI metrics.
Strengths and Weakness of Differing
Diffusion Tensor Imaging Methodologies
As we have previously discussed, the success rate in generating
white matter tracts is marginally higher (after the ACPC
alignment) for Method 2 compared to Method 1. This is likely
because our use of a single subject pseudo atlas restricted
the automated tractography, rendering it more selective in its
ability to generate the white matter tracts. This can be seen
as an advantage to Method 1 as its selectiveness may reduce
the likelihood of generating spurious tracts and thus erroneous
data. The use of Method 2 incorporated the use of validated
Frontiers in Aging Neuroscience | www.frontiersin.org 11 April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 12
Kok et al. Modeling DTI Injury Patterns
FIGURE 5 | DTI radar graph profiles for comparison of methods with and without ACPC alignment and co-registration in Alzheimer’s disease (AD) and cognitively
normal (CN) cohorts. DTI metrics from the ADNI data archive are included for reference. FA values are presented as x20 for illustration; MD, L1, L2, and 3 values are
presented as ×104. gCC, genu of the corpus callosum; UF, uncinate fasciculus; NA, no ACPC alignment.
atlases which have been derived from group-averaging from
a sample cohort. This may also be perceived as providing
this option with a technical advantage over Method 1, which
used a single subject selected from the dataset to generate
the pseudo-atlas. However, as we have discovered, the use
of a pseudo-atlas may equally be argued to be advantageous
as it promotes a template that is more representative of the
cohorts compared to the standardized atlases used in Methods
2 and 3. Our study has shown that this resulted in Method
1 (the pseudo-atlas) being more sensitive than standardized
atlas-based DTI analyses, in characterizing changes in the model
of white matter at-risk due to pathophysiological processes of
distortion and disease.
In terms of processing, Method 1 required a much longer
time to produce the DTI data compared to Methods 2 or 3.
This was for two main reasons. The first was that the white
Frontiers in Aging Neuroscience | www.frontiersin.org 12 April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 13
Kok et al. Modeling DTI Injury Patterns
FIGURE 6 | DTI radar graph profiles for comparison of differences in methods vs. (A) cohort differences and (B) changes in the AD cohort after 12 months. DTI
metrics from the ADNI data archive are included for reference. FA values are presented as ×20 for illustration; MD, L1, L2, and 3 values are presented as ×104. AD,
Alzheimer’s disease; CN, cognitively normal; EMCI, early mild cognitive impairment; gCC, genu of the corpus callosum; UF, uncinate fasciculus.
matter tracts had to be manually generated in the pseudo-atlas
template prior to performing the tractography. This process
could be lengthy and required individuals with a working
knowledge of neuroanatomy to perform. Additionally, there
could be inherent subjectivity when it came to generating
the tracts because it was difficult to determine if there were
Frontiers in Aging Neuroscience | www.frontiersin.org 13 April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 14
Kok et al. Modeling DTI Injury Patterns
FIGURE 7 | White matter tract profiles demonstrating variability along tracts. AD, Alzheimer’s disease; CN, cognitively normal; gCC, genu of the corpus callosum;
CST-R, corticospinal tract (right); UF-R, uncinate fasciculus (right).
missing “strands” of white matter or conversely, if spurious
“strands” were being generated. Secondly, the tractography
itself of Method 1 also required a long time, which required
approximately 3 h to generate a single tract from a single
DWI. In contrast, Methods 2 and 3 did not require manual
generation of the white matter tracts as it utilized readily available
atlases compiled and verified by other groups. In terms of
processing speed, Method 2 was about 50 times faster than
Frontiers in Aging Neuroscience | www.frontiersin.org 14 April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 15
Kok et al. Modeling DTI Injury Patterns
Method 1, requiring 3 h to generate a single tract from a
cohort of 50 DWIs.
Study Limitations
DTI has a low specificity and is generally due to its low signal
to noise ratio (Ranzenberger and Snyder, 2022). As a result, the
imaging quality may be poor, and artifacts become a problem.
Additionally, the DTI metrics are highly dependent on the
size of the voxel during analysis. A single voxel may contain
multi-directional structures which can result in inaccurate DTI
measurements. Ideally, a single voxel should be small enough
that it encompasses a single white matter bundle, taking
a point measurement of DTI metrics. Therefore, the inter-
methodological differences found in our paper could be in part
be attributed to its low signal to noise ratio.
In this study, we only considered two disease cohorts (AD
and EMCI) and healthy controls (CN). The full ADNI dataset
included other cohorts along the disease spectrum, such as the
late mild cognitive impairment (LMCI) and significant memory
concern (SMC) cohorts. With further analysis it may have
been possible that one of the three methods chosen would
have emerged as the predominantly reliable and reproducible
method of DTI analysis, with findings entirely consistent
with literature. In addition, despite widespread use of ROI
methodologies in literature, manual specifications and semi-
automated tractography may be considered less reliable than
fully automated white matter analytical approaches. Nevertheless,
due to their ubiquity, results from this study would be easily
translated to other settings involving DTI analysis at the clinical-
research interface.
In Method 1, a randomly selected image from each study
cohort was used as an atlas. This may have potentially introduced
bias as we could not be certain that the selected images were
adequately representative of the entire cohort. However, the
selected images were inspected for abnormal or outrightly
distinctive features that could render them significantly different
from other images within the cohort. Future work might include
creating a more representative atlas by generating a grouped
average of multiple images from the cohort.
We also recognize that previous studies that have utilized
a representative cohort-specific subject-based approach to
DTI analysis have explored and demonstrated its limitations.
For example, Keihaninejad et al. (2012) compared different
methods of registration schemes for the use of TBSS for DTI
analysis (Standard, Most-Representative-Subject, Study-Specific-
Template, and Group-wise) in terms of their performance
in reducing misalignment within the context of Alzheimer’s
disease and large deformations due to atrophy. They found
that the approaches studied all showed false-positive error
in evaluation of specificity, likely due to variations in levels
of white matter atrophy and ventricular size. However, it
was possible to improve the performance of aligning DTI
data using a group-wise average atlas approach (Keihaninejad
et al., 2012). The degree of confounders such as white matter
atrophy and ventriculomegaly can be highly variable between
patients; it could therefore be equally argued that, in certain
cohorts such as ours, the use of a Most-Representative Subject
approach may still be more advantageous, since we would
expect the white matter pattern changes to affect similar “at-
risk” locations within the same disease process but group-
wise averaging may introduce further unintended distortions
to the template of the “at-risk atlas” of disease. Nevertheless,
our study showed that even in the absence of confounders
such as atrophy and ventriculomegaly as in the CN cohort,
and despite implementation of the algorithmic modifications.
There is still a poor agreement between methodologies, which
supports our conclusion that no true “gold-standard” DTI
methodology exists without limitations for all possible disease
datasets of interest.
It is also important to note that the use of Evans’ Index as a
marker for ventriculomegaly is imperfect because it is dependent
on the inter-rater reliability at measuring the maximal width of
the frontal horns and the internal diameter of the skull. These
measurements are also highly dependent on the chosen slice
and location at which the markers are placed. In addition, the
orientation of each image has a large influence on the slices
and thus the measurements. Although this effect is mitigated
by alignment of the commissures, such technical considerations
should be addressed and optimized by each rater, prior to its
application as a biomarker for ventricular enlargement across a
range of datasets.
Future Work
We plan to expand our analyses using both DTI methodologies,
to include other cohorts of interest along the spectrum of AD
and other neurodegenerative diseases. We also aim to use other
anatomical segmentation methods to examine macro-structural
features of white matter, such as its volume and thickness,
as well as to create topological maps of adjacent surfaces, in
order to augment the interpretation of the morphology of
white matter changes, as described by DTI profiles. In the
context of ventriculomegaly, we plan to utilize complementary
biomarkers for both 2-dimensional and 3-directional measures
in specific groups that possess significant ventriculomegaly such
as cohorts with NPH. Finally, we aim to further expand the
concept of DTI profiles as an invaluable tool toward boosting
our capacity to compare the interpretation of DTI findings
across methodologies which are not directly comparable using
conventional statistical methods.
CONCLUSION
In this study, we found that there was no true gold-standard
for DTI methodologies or atlases. It was possible to create
a pseudo-atlas that was cohort-specific for immediate study
use. Whilst there was no congruence between absolute values
from DTI metrics, differing DTI methodologies were still valid
but must be appreciated to be variably sensitive to different
changes within white matter injury occurring concurrently.
When such changes were found to exist in the same dataset,
the use of differing methods were complementary in elucidating
the characterization of such DTI changes. We found that,
Frontiers in Aging Neuroscience | www.frontiersin.org 15 April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 16
Kok et al. Modeling DTI Injury Patterns
despite such algorithmic modifications, the use of DTI profiles,
a methodology of distilling the complexity of DTI changes to
their most simplistic, graphical forms, confirmed the morphology
of white matter injury as described by DTI metrics, remained
consistent. By combining both atlas and pseudo-atlas based
methodologies with DTI profiles, it was possible to navigate past
such challenges to describe white matter injury changes in the
context of confounders, such as neurodegenerative disease and
ventricular enlargement, with transparency and consistency.
DATA AVAILABILITY STATEMENT
The data will be made available but is subject to approval from
the ADNI (Alzheimer’s Disease Neuroimaging Initiative) as they
own the scans that the data were derived from.
ETHICS STATEMENT
Data used in preparation of this article were obtained from the
ADNI database (adni.loni.usc.edu). Written informed consent
was obtained from all ADNI subjects, and participating sites
in the ADNI study received approval from their respective
governing Institutional Review Boards.
AUTHOR CONTRIBUTIONS
CK, NK, and CL conceptualized and designed the study
methodology, wrote the manuscript, and contributed to the
analysis and interpretation of study data. CK, NK, CL, and
TA contributed to data collection and validation. All authors
contributed to the article and approved the submitted version.
FUNDING
NK was supported by a National Medical Research Council
Clinician Scientist Award (MOH-CSAINV18nov-0005).
ACKNOWLEDGMENTS
We would like to thank Assistant Prof. Seyed Ehsan Saffari for
providing advice on the statistical analyses for this manuscript.
Data used in preparation of this article were obtained from
the Alzheimer’s Disease Neuroimaging Initiative (ADNI)
database (adni.loni.usc.edu). As such, the investigators within
the ADNI contributed to the design and implementation
of ADNI and/or provided data but did not participate in
analysis or writing of this report. A complete listing of ADNI
investigators can be found at: http://adni.loni.usc.edu/wp-
content/uploads/how_to_apply/ADNI_Acknowledgement_List.
pdf. Data collection and sharing for this project was funded
by the Alzheimer’s Disease Neuroimaging Initiative (ADNI)
(National Institutes of Health Grant U01 AG024904) and DOD
ADNI (Department of Defense award number W81XWH-12-2-
0012). ADNI is funded by the National Institute on Aging, the
National Institute of Biomedical Imaging and Bioengineering,
and through generous contributions from the following:
AbbVie, Alzheimer’s Association; Alzheimer’s Drug Discovery
Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-
Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan
Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F.
Hoffmann-La Roche Ltd., and its affiliated company Genentech,
Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer
Immunotherapy Research & Development, LLC.; Johnson
& Johnson Pharmaceutical Research & Development LLC.;
Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics,
LLC.; NeuroRx Research; Neurotrack Technologies; Novartis
Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging;
Servier; Takeda Pharmaceutical Company; and Transition
Therapeutics. The Canadian Institutes of Health Research is
providing funds to support ADNI clinical sites in Canada.
Private sector contributions are facilitated by the Foundation
for the National Institutes of Health (www.fnih.org). The grantee
organization is the Northern California Institute for Research
and Education, and the study is coordinated by the Alzheimer’s
Therapeutic Research Institute at the University of Southern
California. ADNI data are disseminated by the Laboratory for
Neuro Imaging at the University of Southern California.
REFERENCES
Acosta-Cabronero, J., Alley, S., Williams, G. B., Pengas, G., and
Nestor, P. J. (2012). Diffusion tensor metrics as biomarkers in
Alzheimer’s disease. PLoS One 7:e49072. doi: 10.1371/journal.pone.00
49072
Acosta-Cabronero, J., and Nestor, P. J. (2014). Diffusion tensor imaging
in Alzheimer’s disease: insights into the limbic-diencephalic network and
methodological considerations. Front. Aging Neurosci. 6:266. doi: 10.3389/fnagi.
2014.00266
Assaf, Y., Ben-Sira, L., Constantini, S., Chang, L. C., and Beni-Adani, L. (2006).
Diffusion tensor imaging in hydrocephalus: initial experience. AJNR Am. J.
Neuroradiol. 27, 1717–1724.
Ben-Sira, L., Goder, N., Bassan, H., Lifshits, S., Assaf, Y., and Constantini,
S. (2015). Clinical benefits of diffusion tensor imaging in hydrocephalus.
J. Neurosurg. Pediatr. 16, 195–202. doi: 10.3171/2014.10.PEDS
13668
Borden, N. M., Forseen, S. E., and Stefan, C. (2015). Imaging Anatomy of the
Human Brain : A Comprehensive Atlas Including Adjacent Structures. New York,
NY: Springer Publishing Company.
Christidi, F., Karavasilis, E., Samiotis, K., Bisdas, S., and Papanikolaou, N. (2016).
Fiber tracking: a qualitative and quantitative comparison between four different
software tools on the reconstruction of major white matter tracts. Eur. J. Radiol.
Open 3, 153–161. doi: 10.1016/j.ejro.2016.06.002
Fedorov, A., Beichel, R., Kalpathy-Cramer, J., Finet, J., Fillion-Robin, J. C., Pujol,
S., et al. (2012). 3D slicer as an image computing platform for the quantitative
imaging network. Magn. Reson. Imaging 30, 1323–1341. doi: 10.1016/j.mri.
2012.05.001
Glasser, M. F., Sotiropoulos, S. N., Wilson, J. A., Coalson, T. S., Fischl, B.,
Andersson, J. L., et al. (2013). The minimal preprocessing pipelines for
the human connectome project. Neuroimage 80, 105–124. doi: 10.1016/j.
neuroimage.2013.04.127
Hattori, T., Yuasa, T., Aoki, S., Sato, R., Sawaura, H., Mori, T., et al. (2011).
Altered microstructure in corticospinal tract in idiopathic normal pressure
Frontiers in Aging Neuroscience | www.frontiersin.org 16 April 2022 | Volume 14 | Article 787516

fnagi-14-787516 April 21, 2022 Time: 14:29 # 17
Kok et al. Modeling DTI Injury Patterns
hydrocephalus: comparison with Alzheimer disease and Parkinson disease with
dementia. AJNR Am. J. Neuroradiol. 32, 1681–1687. doi: 10.3174/ajnr.A2570
Hofer, S., and Frahm, J. (2006). Topography of the human corpus callosum
revisited–comprehensive fiber tractography using diffusion tensor magnetic
resonance imaging. Neuroimage 32, 989–994. doi: 10.1016/j.neuroimage.2006.
05.044
Kanno, S., Abe, N., Saito, M., Takagi, M., Nishio, Y., Hayashi, A., et al. (2011). White
matter involvement in idiopathic normal pressure hydrocephalus: a voxel-based
diffusion tensor imaging study. J. Neurol. 258, 1949–1957. doi: 10.1007/s00415-
011-6038- 5
Keihaninejad, S., Ryan, N. S., Malone, I. B., Modat, M., Cash, D., Ridgway, G. R.,
et al. (2012). The importance of group-wise registration in tract based spatial
statistics study of neurodegeneration: a simulation study in Alzheimer’s disease.
PLoS One 7:e45996. doi: 10.1371/journal.pone.0045996
Keong, N. C., Pena, A., Price, S. J., Czosnyka, M., Czosnyka, Z., Devito, E. E., et al.
(2017). Diffusion tensor imaging profiles reveal specific neural tract distortion
in normal pressure hydrocephalus. PLoS One 12:e0181624. doi: 10.1371/journal.
pone.0181624
Lebel, C., Walker, L., Leemans, A., Phillips, L., and Beaulieu, C. (2008).
Microstructural maturation of the human brain from childhood to adulthood.
Neuroimage 40, 1044–1055. doi: 10.1016/j.neuroimage.2007.12.053
Leemans, A. J., Jeurissen, B., Sijbers, J., and Jones, D. K. (2009). “ExploreDTI: a
graphical toolbox for processing, analyzing, and visualizing diffusion MR data,”
in in: Proceedings of the 17th Annual Meeting of International Societyof Magnetic
Resonance Medicine, (Hawaii), 3537.
Lock, C., Kwok, J., Kumar, S., Ahmad-Annuar, A., Narayanan, V., Ng, A. S. L.,
et al. (2018). DTI profiles for rapid description of cohorts at the clinical-research
interface. Front. Med. (Lausanne) 5:357. doi: 10.3389/fmed.2018.00357
Mori, S., and Zhang, J. (2006). Principles of diffusion tensor imaging and its
applications to basic neuroscience research. Neuron 51, 527–539. doi: 10.1016/
j.neuron.2006.08.012
Mukherjee, P., Chung, S. W., Berman, J. I., Hess, C. P., and Henry, R. G. (2008).
Diffusion tensor MR imaging and fiber tractography: technical considerations.
AJNR Am. J. Neuroradiol. 29, 843–852. doi: 10.3174/ajnr.A1052
Nir, T. M., Jahanshad, N., Villalon-Reina, J. E., Toga, A. W., Jack, C. R., Weiner,
M. W., et al. (2013). Effectiveness of regional DTI measures in distinguishing
Alzheimer’s disease. MCI, and normal aging. Neuroimage Clin. 3, 180–195.
doi: 10.1016/j.nicl.2013.07.006
(Nitrc) N.T.R.C (2014). MRIcroGL. Available Online at (https://www.nitrc.org/
projects/mricrogl/) (accessed March 2, 2021).
Oishi, F., Van Zijl, P. C. M., Mori, S., and Andreia, V. (2010). MRI Atlas of Human
White Matter. Cambridge, MS: Academic Press.
Oishi, K., Faria, A., Jiang, H., Li, X., Akhter, K., Zhang, J., et al. (2009).
Atlas-based whole brain white matter analysis using large deformation
diffeomorphic metric mapping: application to normal elderly and Alzheimer’s
disease participants. Neuroimage 46, 486–499. doi: 10.1016/j.neuroimage.2009.
01.002
Oishi, K., Zilles, K., Amunts, K., Faria, A., Jiang, H., Li, X., et al. (2008). Human
brain white matter atlas: identification and assignment of common anatomical
structures in superficial white matter. Neuroimage 43, 447–457. doi: 10.1016/j.
neuroimage.2008.07.009
Paul, D. A., Gaffin-Cahn, E., Hintz, E. B., Adeclat, G. J., Zhu, T., Williams,
Z. R., et al. (2014). White matter changes linked to visual recovery after
nerve decompression. Sci. Transl. Med. 6:266ra173. doi: 10.1126/scitranslmed.
3010798
R Development Core Team. (2010). R: A language and Environment for Statistical
Computing". 4.0, 4 Edn. Vienna: R Foundation for Statistical Computing.
Ranzenberger, L. R., and Snyder, T. (2022). Diffusion Tensor Imaging. Treasure
Island, FL: StatPearls Publishing.
Scheel, M., Diekhoff, T., Sprung, C., and Hoffmann, K. T. (2012). Diffusion
tensor imaging in hydrocephalus–findings before and after shunt surgery. Acta
Neurochir. (Wien) 154, 1699–1706. doi: 10.1007/s00701-012-1377-2
Smith, S. M., Jenkinson, M., Johansen-Berg, H., Rueckert, D., Nichols, T. E.,
Mackay, C. E., et al. (2006). Tract-based spatial statistics: voxelwise analysis of
multi-subject diffusion data. Neuroimage 31, 1487–1505.
Soares, J. M., Marques, P., Alves, V., and Sousa, N. (2013). A hitchhiker’s guide
to diffusion tensor imaging. Front. Neurosci. 7:31. doi: 10.3389/fnins.2013.
00031
Soon, S. X. Y., Kumar, A. A., Tan, A. J. L., Lo, Y. T., Lock, C., Kumar, S., et al. (2021).
The impact of multimorbidity burden, frailty risk scoring, and 3-directional
morphological indices vs. testing for csf responsiveness in normal pressure
hydrocephalus. Front Neurosci. 15:751145. doi: 10.3389/fnins.2021.751145
Tan, K., Meiri, A., Mowrey, W. B., Abbott, R., Goodrich, J. T., Sandler, A. L.,
et al. (2018). Diffusion tensor imaging and ventricle volume quantification
in patients with chronic shunt-treated hydrocephalus: a matched case-control
study. J. Neurosurg. 129, 1611–1622. doi: 10.3171/2017.6.JNS162784
Wakana, S., Caprihan, A., Panzenboeck, M. M., Fallon, J. H., Perry, M., Gollub,
R. L., et al. (2007). Reproducibility of quantitative tractography methods applied
to cerebral white matter. Neuroimage 36, 630–644. doi: 10.1016/j.neuroimage.
2007.02.049
Yamada, S., Ishikawa, M., and Yamamoto, K. (2016). Comparison of CSF
distribution between idiopathic normal pressure hydrocephalus and Alzheimer
Disease. AJNR Am. J. Neuroradiol. 37, 1249–1255. doi: 10.3174/ajnr.A4695
Yeatman, J. D., Dougherty, R. F., Myall, N. J., Wandell, B. A., and Feldman,
H. M. (2012). Tract profiles of white matter properties: automating fiber-tract
quantification. PLoS One 7:e49790. doi: 10.3389/fnins.2013.00031
Yendiki, A., Panneck, P., Srinivasan, P., Stevens, A., Zollei, L., Augustinack, J.,
et al. (2011). Automated probabilistic reconstruction of white-matter pathways
in health and disease using an atlas of the underlying anatomy. Front.
Neuroinform. 5:23. doi: 10.3389/fninf.2011.00023
Zalesky, A. (2011). Moderating registration misalignment in voxelwise
comparisons of DTI data: a performance evaluation of skeleton projection.
Magn. Reson. Imaging 29, 111–125. doi: 10.1016/j.mri.2010.06.027
Conflict of Interest: The authors declare that the research was conducted in the
absence of any commercial or financial relationships that could be construed as a
potential conflict of interest.
Publisher’s Note: All claims expressed in this article are solely those of the authors
and do not necessarily represent those of their affiliated organizations, or those of
the publisher, the editors and the reviewers. Any product that may be evaluated in
this article, or claim that may be made by its manufacturer, is not guaranteed or
endorsed by the publisher.
Copyright © 2022 Kok, Lock, Ang and Keong. This is an open-access article
distributed under the terms of the Creative Commons Attribution License (CC BY).
The use, distribution or reproduction in other forums is permitted, provided the
original author(s) and the copyright owner(s) are credited and that the original
publication in this journal is cited, in accordance with accepted academicpractice. No
use, distribution or reproduction is permitted which does not comply with theseterms.
Frontiers in Aging Neuroscience | www.frontiersin.org 17 April 2022 | Volume 14 | Article 787516