REVIEW ARTICLE
Beyond Diffusion Tensor MRI Methods for Improved
Characterization of the Brain after
Ischemic Stroke: A Review
E.V.R. DiBella,
A. Sharma,
L. Richards,
V. Prabhakaran,
J.J. Majersik, and
S.K. HashemizadehKolowri
ABSTRACT
SUMMARY: Ischemic stroke is a worldwide problem, with 15 million people experiencing a stroke annually. MR imaging is a valuable
tool for understanding and assessing brain changes after stroke and predicting recovery. Of particular interest is the use of diffusion MR imaging in the nonacute stage 1–30 days poststroke. Thousands of articles have been published on the use of diffusion MR
imaging in stroke, including several recent articles reviewing the use of DTI for stroke. The goal of this work was to survey and put
into context the recent use of diffusion MR imaging methods beyond DTI, including diffusional kurtosis, generalized fractional anisotropy, spherical harmonics methods, and neurite orientation and dispersion models, in patients poststroke. Early studies report
that these types of beyond-DTI methods outperform DTI metrics either in being more sensitive to poststroke changes or by better
predicting outcome motor scores. More and larger studies are needed to confirm the improved prediction of stroke recovery with
the beyond-DTI methods.
ABBREVIATIONS: AD ¼ axial diffusivity; AK ¼ axial kurtosis; DKI ¼ diffusional kurtosis imaging; FA ¼ fractional anisotropy; FM ¼ Fugl-Meyer; GFA ¼ generalized fractional anisotropy; MD ¼ mean diffusivity; MK ¼ mean kurtosis; NODDI ¼ neurite orientation dispersion and density imaging; PLIC ¼ posterior limb
of the internal capsule; RD ¼ radial diffusivity; SHORE ¼ simple harmonic oscillator-based reconstruction and estimation; vic ¼ neurite density; WMTI ¼ white
matter tract integrity
S
troke or a cerebrovascular accident is a problem worldwide,
with 15 million people having a stroke annually, and it is a
widespread cause of long-term disability and mortality.1 MR
imaging is a valuable tool for understanding and assessing brain
changes after stroke and predicting recovery. Of particular interest is the use of diffusion MR imaging after the hyperacute stage,
at 1–30 days poststroke. Thousands of articles have been published on the use of diffusion MR imaging in stroke, including
several recent articles reviewing the use of DTI in stroke.2-4 The
goal of this work was to survey and put into context the recent
use of diffusion MR imaging methods beyond DTI, including diffusional kurtosis imaging (DKI),5 generalized fractional anisotropy (GFA),6 and neurite orientation dispersion and density
imaging (NODDI) models,7 in patients poststroke. These methods use functional representations that either better match wider
Received September 6, 2021; accepted after revision November 8.
From the Departments of Radiology and Imaging Sciences (E.V.R.D., A.S., S.K.H.),
Occupational and Recreational Therapies (L.R.), and Neurology (J.J.M.), University of
Utah, Salt Lake City, Utah; and Department of Radiology (V.P.), University of
Wisconsin, Madison, Wisconsin.
Please address correspondence to Ed DiBella, PhD, UCAIR/Radiology and Imaging
Sciences, 729 Arapeen Dr, Salt Lake City, UT 84108; e-mail:
Edward.dibella@hsc.utah.edu
Indicates open access to non-subscribers at www.ajnr.org
Indicates article with online supplemental data.
http://dx.doi.org/10.3174/ajnr.A7414
ranges of diffusion data or tie to biophysical models that may better inform stroke studies. Early works report that these types of
methods outperform DTI metrics.
Time Course of Ischemic Stroke
Approximately 87% of strokes are ischemic, while 13% are hemorrhagic; this review focuses on ischemic stroke. Because more ischemic strokes occur in the territory of the MCA than in other
locations, most ischemic strokes affect the motor system. For
example, 80% of individuals with stroke have impaired upper extremity motor function acutely.8,9
The typical time course of ischemic stroke (Online
Supplemental Data) begins with an acute phase in which ischemia results from development of either an in situ or embolic
thrombus that lodges in a cerebral blood vessel, reducing or
stopping blood flow to neural tissue served by that vessel. Soon
after reduction of downstream blood flow, cytotoxic edema
develops in the ischemic area and the cells in both gray and
white matter swell, reducing the extracellular space from 20%
to 4%–10%.10 Such cell swelling is due to the lack of oxygen that
impedes adenosine triphosphate production needed for active
transport by Na1/K1 ATPase to keep Na ions in balance.10
Ischemic areas with blood flow of , 10mL/100g become damaged in ,6 minutes,10 creating an ischemic core.11 Membrane
configurations also change and form blebs,10,12 also termed
AJNR Am J Neuroradiol 43:661–69
May 2022 www.ajnr.org
661
axonal beading.13 Acute treatments with tPA or catheter-based
clot removal can be very helpful in limiting or avoiding neural
damage if the patient presents early poststroke.
During the next weeks and months, the ischemic area turns
into a necrotic core and the whole brain changes in a variety of
ways. In the peri-ischemic area, Wallerian degeneration14 of the
myelinated tracts is generally detectable at a few days poststroke
and continues for months. Wallerian degeneration is the process
of demyelination and disintegration of the distal axons that
occurs after injury to a neuron. Axonal swelling is also characteristic of the early stage of Wallerian degeneration. Wallerian
degeneration may occur both near and far from the ischemic bed.
For example, rats with MCA occlusion exhibited Wallerian
degeneration in the nonischemic corpus callosum.14
Lesions in one hemisphere also promote changes in the unlesioned (contralateral) hemisphere, depending on the amount of
damage to the ipsilesional motor system. Degeneration of transcallosal fibers, increased branching and pruning of dendrites, and
increased contralateral corticospinal tract sprouting have been
reported.15 Jones and Adkins16 suggest that overcompensation
with the nonparetic side promotes remodeling in the contralesional hemisphere and may cause synaptic competition with the
lesioned hemisphere and reduce recovery in the paretic limb.
Neuroimaging with MR imaging plays several roles in stroke;
the focus of this review is the use of diffusion MR imaging in the
nonhyperacute-but-early stage 1–30 days poststroke, sometimes
termed the subacute stage.
Diffusion MR imaging is widely used to assess and study
stroke. MR imaging diffusion-weighted images reflect how water
protons diffuse during short time spans. The water diffusivity
depends on the local environment of the water protons. For
example, in intact myelinated fibers, water protons diffuse with
little restriction along the axon or fiber. In CSF, diffusion is isotropic and protons move relatively freely in all directions. While
MR imaging diffusion images are acquired on millimeter scales,
the diffusion reflects hindrances and restrictions to water mobility due to microstructures, with scales on the order of microns.
Thus, diffusion provides information about the integrity of neural
microstructures, but in a complicated way because the brain has a
variety of neuronal cell bodies, axons, dendrites, and glial cells in
each image voxel. The different compartments and the orientations of the fibers in the cell and the intra- and extracellular
spaces influence the diffusion signal.17
Standard diffusion MR imaging acquires unweighted (b ¼ 0)
and diffusion-weighted images with diffusion weighting in multiple directions. Parameters such as ADC maps are calculated. The
diffusion tensor signal representation uses a low-order cumulant
expansion of diffusion18 and enables obtaining parameters such
as fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD). Diffusion imaging and
DTI have been reviewed in many places.19 The Online
Supplemental Data show an example of some of these different
parameters from DTI.
There has also been a tremendous amount of work, which is
doubling every 2.9 years (see Fig 1 in Novikov et al20), on diffusion MR imaging microstructural mapping techniques. These
methods use higher-order signal expansions (such as DKI) or
662
DiBella May 2022
www.ajnr.org
other signal representations21 or model-based approximations to
the biophysics. The motivation is that these models can be more
specific to particular structures in a voxel and thus give more information regarding health or disease. The different signal representations and models are derived in different ways and require
different data acquisitions than DTI, in particular with regard to
q-space sampling and b-values (Online Supplemental Data).
However, many of the parameters can be derived from the same
acquisitions, and some of the studies listed in the Online
Supplemental Data give parameters from several different representations or models, using the same data set. The Online
Supplemental Data also show an example of some of these different parameters from most of the methods that are detailed later
in this review.
For stroke, most of the work has focused on the corticospinal tract and white matter.22 Gray matter has been studied in
healthy individuals with a NODDI model23 and with a model
designed for gray matter 24 (which may require very high b-values). There are limited studies using beyond-DTI microstructural mapping methods to look at gray matter in human
subjects with stroke, though DKI5 and diffusion spectrum
imaging25 have been used.
Diffusion MR Imaging in Stroke
In regions of ischemic stroke, signal in diffusion-weighted images
is decreased acutely.26 The reason that diffusion decreases in
stroke is not completely understood, though a number of studies
give some insight into the process. For example, as described
above, ischemic neurons undergo axonal swelling. If axons swell
and intra-axon diffusion compared with extra-axonal diffusion is
quite different, then having more intra-axonal space, which is
more restricted to diffusion, could result in lower diffusion (and
higher signal) on the MR images. However, a study using fluorodeoxyglucose molecules found that the molecule diffuses equally
in both intra- and extracellular spaces and diffusion in both
spaces decreases (40% ADC decrease) in ischemic conditions.27
More recent works have evolved estimates of intra-axonal diffusion and its relation to extracellular diffusion.28-30 A likely partial
explanation for slower, more restricted diffusion in stroke regions
is that axons become “beaded,” more tortuous instead of purely
cylindric structures so that the total cell surface area is preserved
during the swelling.31 In this case, diffusion may become more
restricted, especially in the axonal direction. Simulations have
shown that water ADC decreases in beaded structures.13,31
Photomicrographs12 and confocal microscopy have shown such
structures.32
Other factors such as the multiexponential diffusion change
with b-value, water exchange, and the axonal g-ratio (outer-toinner axon diameter ratio),20 also play roles in imaging diffusion
after stroke. Water exchange between fast and slow diffusion
components has been studied,33 and the biophysics of white matter in the context of diffusion imaging was reviewed by Nilsson et
al.34 More recently, a beyond-DTI method sensitive to exchange
was proposed (Online Supplemental Data).35
We move now to consider some of the works that illustrate
how MR imaging has been used in the assessment of stroke and
for prediction of stroke recovery.
Diffusion MR Imaging for Predicting Recovery from Stroke.
Diffusion MR imaging has many potential uses in subacute
stroke, including understanding the time course of brain changes,
predicting the response to specific therapeutic interventions, and
selecting and monitoring therapy. Predicting motor recovery has
been a highly studied objective.2
Why Predict Recovery? Stroke research includes both hyperacute
evaluation and treatment with thrombolysis or mechanical clot
removal and less acute evaluation. “Time is brain,” and as
described above, and for suitable patients presenting to the hospital within 24 hours, tPA and thrombectomy procedures often
help recovery by reducing the extent of neural damage. Yet, while
these procedures reduce damage from stroke, they usually do not
eliminate such damage. After hyperacute procedures and for the
80% of subjects who do not receive such intervention, rehabilitation is commonly needed. How the brain recovers after stroke
and the best therapy and therapy timing for each stroke survivor
after the hyperacute phase are open questions. Therapy within
the first few weeks after stroke is likely to be more effective due to
the heightened neuroplasticity potential at that time.
Predicting recovery is, thus, a highly sought-after goal.36-38
Accurate prediction is needed to plan discharge, set patient goals,
choose therapy (therapies aimed at reducing impairment versus
therapies to train compensation strategies and assistive technologies), and offer the patient a prognosis.39 Recovery predictions
also allow better testing of newly developed recovery therapies
that hope to promote greater recovery than spontaneous recovery
or current therapies.
Nonimaging Methods for Stroke Recovery Prediction. A variety
of prediction methods have been used, such as those based on
changes in the NIHSS,40 but prediction methods that do not use
imaging typically have poor specificity (36% in Hemmen et al40)
or only estimate a binary outcome (favorable or not favorable),
which is not as useful clinically. Recovery depends, in part, on the
size and the location of the ischemic area and on a variety of
other factors such as demographics, systemic atherosclerosis, and
social support networks.
Strokes causing motor deficits are commonly said to display
proportional recovery—that is, in the first 3 months following
stroke, 70% of the new impairment is recovered41-44 in most individuals, while the other individuals (“atypical responders”) have
much less recovery. This finding is based on function at baseline
measured with the Fugl-Meyer (FM) upper extremity assessment,45 which has a range of 0–66. For example, DFM = 0.7
(66-FMinitial) 1 0.4 predicted recovery reasonably well in 160 of
211 subjects.43 However, it was recently shown by two separate
groups that mathematic coupling and ceiling effects incorrectly
biased the previous findings and that random recovery could give
results similar to proportional recovery when the previously published analysis methods were used.46,47 Thus, we may understand
recovery less than we think we do, and there is additional motivation for methods such as MR imaging to bolster our ability to
characterize and predict recovery from stroke.
Imaging Methods for Prediction of Stroke Recovery. MR imaging methods such as diffusion, perfusion, fMRI, spectroscopy,
and T1- and T2-weighted images give insight into the location
and severity of stroke.48,49 Perfusion imaging to delineate ischemic areas is typically used in concert with diffusion imaging at
the hyperacute stage of stroke to see if the ischemic area exceeds
(defining a larger area at risk) or matches the restricted diffusion
area. Match/mismatch criteria can inform the hyperacute use of
thrombectomy and tPA.50 These methods are applied at the
hyperacute stage and are typically less valuable for longer-term
recovery prediction. fMRI was shown to relate more weakly to
motor deficits than tract integrity based on diffusion imaging.51
Moreover, task fMRI relies on patient cooperation, effort, and
ability to perform motor tasks, which may be compromised on
the basis of the severity of stroke, though resting state functional
connectivity may be promising.52 T1 and T2 methods are often
used to size lesions but correlate relatively poorly with motor
function compared with also considering lesion location and load
on a fiber tract (gauged with diffusion imaging53,54). Lesion load
is also a relatively poor predictor of recovery. Adding DTI enables
better assessment of stroke and prediction of recovery.
DTI Methods for Predicting Stroke Recovery
DTI uses a signal representation that is limited to representing,
for example, a Gaussian distribution of diffusion displacement
and a single fiber direction in each voxel, which may not accurately reflect the mixture of glia, myelin, neurons, extracellular
space, CSF, and microvasculature that can exist within a voxel.
Still, DTI metrics are sensitive to stroke-related changes, and
numerous studies have used DTI to better understand and characterize changes after stroke48,54 and to gauge the effects of
therapies.55
While DTI and particularly FA are useful, histology studies
have shown that FA is not a specific biomarker because it conflates myelination, fiber dispersion, and intra-/extraneurite contributions. FA is widely used despite its nonspecific nature. For
example, a recent review4 reported good correlation between a
DTI metric such as FA and a motor outcome metric such as FM
or NIHSS (which is not as useful a measure of motor capabilities
as FM) scores, though a few of the studies did not find significant
predictive correlations from the FA metrics at some stages.56-58
Kumar et al2 did a meta-analysis on 6 small studies with outcome
measures such as NIHSS and reported a pooled r = 0.82. They
pointed out the need for larger, prospective studies to better
determine the utility of FA metrics for predicting outcomes. As
highlighted below, these retrospective studies report best-fit linear correlations for the data obtained and would not be expected
to perform as well when doing prospective predictions.
Tractography from DTI has also been useful for studying
stroke and predicting recovery. For example, tractography methods for defining tract-based ROIs4 and connectivity analysis
methods59,60 have been developed.
Diffusion methods stand to benefit from beyond-DTI methods that can in some ways consider crossing fibers, structural
compartments with hindered or restricted diffusion, and nonGaussian distributions of the diffusive motion of water in the
brain. Here, we seek to complement the recent DTI in stroke
reviews2-4 by focusing on works that used beyond-DTI methods
for obtaining microstructural parameter maps in stroke. Several
AJNR Am J Neuroradiol 43:661–69 May 2022
www.ajnr.org
663
studies have reported better correspondence of these microstructural mapping methods with motor outcomes, and others have
reported greater changes between ipsilesional and contralesional
regions in the microstructural parameters compared with DTI,
implying greater sensitivity to stroke effects.
Beyond-DTI Methods for Stroke
New diffusion methods have been developed that better separate
the diffusion signal contributions from fiber dispersion and
intra-/extraneurite compartments. More samples in q-space
(Online Supplemental Data) provide better angular resolution of
fibers,61-65 allow estimating the non-Gaussian portions of the signals (kurtosis), and enable estimating models of biophysical compartments along with fiber orientations and dispersions. Such
compartment models are well-described in multiple articles,
including the supplement in Lakhani et al.66 Because the adverse
impact of stroke is, in part, from disrupting white matter connectivity, better measures of assessing fiber bundle integrity could
give better insight and predictive power over current techniques.
This white matter assessment is one way that beyond-DTI methods may provide significant additional information. Currently,
the only beyond-DTI methods used with stroke applications to
date include the following: DKI, white matter tract integrity
(WMTI), GFA, NODDI, simple harmonic oscillator-based reconstruction and estimation (SHORE) parameters, and 1 article
using different diffusion times to estimate the rate of kurtosis.
DKI. The standard DTI representation is appropriate in freely diffusing isotropic media such as CSF. For areas with restricted diffusion such as within axons, Jensen et al67 proposed extending
the DTI signal representation to add a higher-order term:
SðbÞ
lnð
Þ¼
Sð0Þ
1
bD þ b2 D2 K;
6
where K is the kurtosis. To have sufficient data to fit these higherorder terms uniquely, one must have data acquired at multiple
nonzero b-values. As with DTI, a kurtosis tensor can be defined
to take into account diffusion directions. When DKI is used to fit,
the DTI tensor portion has 6 unknowns and the DKI tensor values include 15 unknowns, so at least 21 q-space samples and a
b=0 image are typically needed. For typical brain microstructure
and diffusion acquisition parameters, the non-Gaussian portion
of the diffusion signal68 will not be prevalent if b#1000 mm2/s.
Acquisitions vary but may include, for example, 15 directions at
b=1000 and 15 directions at b=2000 (Online Supplemental Data).
The additional microenvironment complexity in stroke may
make DKI a better approach than DTI.
Most of the beyond-DTI research in stroke has been with
DKI, with 10 articles in rat stroke models,69-78 a recent review article,79 and 10 articles in humans (Online Supplemental Data).
The rat model studies found larger changes with mean kurtosis
(MK) and axial kurtosis (AK) compared with DTI metrics. Sizing
of lesions was also studied with ex vivo MR imaging, and the
studies found that DKI metrics such as MK were larger than
those in MD and histology.72,78 However, with in vivo MR imaging and a different segmentation method, others reported that
664
DiBella May 2022
www.ajnr.org
the DKI metrics yielded lesion sizes smaller than MD volumes.77
In addition, sizing has not been shown to be very predictive of
motor recovery.80 Possibly more promising from the rat model
studies is that the DKI parameters changed more with time (or
compared with the contralateral hemisphere) than the DTI parameters due to stroke. This increased sensitivity to stroke was
also found in the studies with humans. The Online Supplemental
Data summarize the details of each of these DKI human imaging
poststroke articles.
The relatively widespread use of DKI compared with other
beyond-DTI methods is likely due to modest acquisition requirements, GE Healthcare’s early implementation of DKI in their
FuncTool software on the scanner, and early reports of potential.
For example, Hui et al,5 in 2012, reported that MK changed more
than FA in stroke regions relative to the nonstroke hemisphere
(ipsilesional and contralesional regions) in 44 subjects with
stroke, though there was no evaluation of the clinical importance
of such changes. Fast kurtosis methods using ,20 diffusionweighted images have also been developed81 and used in rat
studies.71,74
Ideally, FM scores at 6–12 weeks poststroke serve to measure
motor outcomes, and such scores were measured in 2 DKI
articles. In Spampinato et al,82 the MK ipsilesional/contralesional
ratio gave the best correlation with FM scores (r = 0.85). The AK
ratio also was predictive (r = 0.78), while DTI MD (r = 0.69), and
FA (r = 0.4) ratios had lower associations with FM scores. Li et
al83 grouped patients into good responders (.10-point FM
change) or poor responders. Then a relative axial kurtosis AK
measure (ipsilesional–contralesional)/contralesional) gave the
best discrimination of the 2 groups and was superior to DTI
methods. The study in Yu et al84 used NIHSS scores rather than
FM and found lower correlations, but still the relative AK metric
gave the highest correlation (r = 0.3) with NIHSS scores. DTI
methods in Yu et al showed no changes in the cortical spinal tract
on the stroke side relative to the contralateral side (and hence no
correlation with NIHSS).
The WMTI method was included in 2 of the DKI articles.5,82
WMTI assumes intra- and extra-axonal compartments and estimates axonal water fraction, intra-axonal AD, extra-axonal axial
and radial diffusivities, and the tortuosity of the extra-axonal
space. WMTI assumes all parallel axons and so is only applied in
the corpus callosum, or where FA .0.3.5,82 WMTI parameters
were calculated along with DKI, and the axonal water fraction
correlated (r = 0.63) with FM scores in Spampinato et al.82 More
recent nonstroke works relaxed the parallel axon requirement
using a WMTI-Watson model that includes fiber dispersion,
though the method assumes axially symmetric diffusion, so it was
only applied in the spinal cord.85
Besides DKI, other beyond-DTI methods have been explored
for use in patients with stroke. The Online Supplemental Data
summarize the 9 articles that have investigated other methods:
GFA, SHORE, NODDI, and rate of kurtosis methods have all
shown promising results in patients with stroke.
GFA. GFA is obtained from the orientation distribution function,
which can be calculated from the 3D ensemble average propagator.21 The 3D ensemble average propagator is obtained, for
example, by 3D Fourier transform from Cartesian-sampled qspace data.86 Then, similar to how FA is calculated from DTI
data as SD(l ) / RMS(l ), where l is the eigenvalue of the diffusion tensor and RMS stands for root mean square, GFA ¼ SD
( w )/RMS( w ), where w represents the orientation distribution
function.21 GFA can be considered as an extension of FA in voxels that have crossing/touching fibers, so it may give a better measure of fiber disarray in stroke.
The GFA studies in the Online Supplemental Data came from
diffusion spectrum imaging, which acquires a Cartesian grid of qspace samples with a relatively high bmaximum (4000–8000 s/mm2).
Most of these studies took the unusual approach of studying only
the contralesional side. GFA revealed microstructure changes in
the contralesional side during poststroke remodeling.6,25,87-89 Most
of these studies used the same data set of 10 patients with stroke
who had been imaged with a 26-minute diffusion acquisition.6,25,89
GFA in the contralesional side was highly predictive of the 6 month NIHSS score (r2 ¼ 0.84 adjusted).6 Combining GFA 1
NIHSS1 age had an even higher correlation (r2 ¼ 0.96 adjusted)
in this small study.6 Other parameters (SHORE) were also calculated from this data set, as described below.
SHORE. The SHORE method calculates propagator anisotropy,
along with measures of the ensemble average propagator variance
(mean square displacement), return-to-the-origin probability, the
return-to-the-axis probability, and the return-to-the-plane probability.90 These parameters when evaluated in a subcortical loop of
tracts gave high correlations with 6-month NIHSS scores.91
Galazzo et al25 added comparison with DTI and evaluated findings in gray matter. Both articles used the 10-person diffusion
spectrum imaging data set from Granziera et al6 and evaluated
only the contralesional side.
Compartment Model Methods. Unlike the FA, GFA, and
SHORE parameters that arise from signal representations that do
not postulate a particular underlying biophysical model, a number of compartment-based models have been developed. These
models, of which NODDI is a good example, estimate microstructural multicompartment (intra-axon, extra-axon, CSF compartments for NODDI) and fiber-dispersion parameters. This
approach is thought to be useful for characterizing dispersed or
crossing fibers92 and for estimating white matter degeneration in
the subacute phase of stroke when Wallerian degeneration and
glial scarring processes are occurring.93 A few groups have
recently reported NODDI models to be useful in stroke imaging
in rats94 and humans.7,95-97 These findings are despite the fact
that NODDI fixes diffusivity parameters to 1.7 (and to 3.0 s/mm2
in spinal fluid), and these parameters are not uniformly correct in
healthy individuals and are less accurate in stroke regions. It is
also known that NODDI “neurite density” (also called vic) does
not reflect the density of neurites in tumor and in stroke (where
it has been called the “restricted diffusion index”). The only
NODDI study including outcome measures (FM scores) was
Hodgson et al,96 in which the best outcome predictor was found
to be the orientation dispersion index (optimism-adjusted r2 ¼
0.83), though vic (or the “restricted diffusion index”) also
correlated r2 ¼ 0.70, as did GFA (r2 ¼ 0.57). Other parameters
including DTI and stroke size and lesion load did not correlate as
well.
The two other NODDI stroke studies included DKI and also
studied changes in microstructure across time.7,97 The study of
Wang et al 97 was unpaired: The same subject was not imaged at
multiple time points. Still, the percentage difference of the orientation dispersion index in the stroke area relative to the contralesional side was larger than other NODDI parameters, MK, FA, or
MD. Correlation with time since stroke onset was also highest
with the orientation dispersion index. Mastropietro et al7 analyzed only the posterior limb of the internal capsule (PLIC) and
cerebral peduncle regions and reported significant differences
between their ipsilateral and contralateral sides for the orientation
dispersion index and other parameters including FA, AD, RD,
kurtosis anisotropy, and AK. Other parameters (MD, MK, vic,
isotropic volume fraction) did not show significant differences.
Other biophysical factors, such as different T2s in intra-axonal
and extracellular compartments, water exchange, and Gaussian
diffusion assumed in compartments, could all be of interest but
are not included in the compartment model methods. Using different diffusion times while maintaining the same b-value is also
of interest and was recently shown to help predict chronic stroke
areas in 5 subjects with stroke.35 That method estimated the
exchange rate (rate of kurtosis) using an equation based on a 2compartment model with exchange. Other models such as
NODDI assume no water exchange between the myelin and the
intracellular and extracellular compartments, though there is evidence for a mixture of fast and slow exchange.34 Such exchange
causes signal changes due to different diffusivities and relaxivities
(possibly negligible) in the different compartments.98 A multishell acquisition with 2 diffusion times was used in Lampinen et
al35 to obtain a parameter sensitive to the exchange rate. Results
showed faster exchange in the ischemic stroke areas.35
The Online Supplemental Data summarize the methods with
outcome measures (FM or NIHSS) graphically, with the graph
edges reflecting the number of subjects in each comparison.
Acquisition and Analysis Diversity in Beyond-DTI Methods
The beyond-DTI studies varied considerably in the acquisition.
Besides a range of angular q-space sampling and different shells or
b-values, there were wide differences in spatial resolution (0.94–
3 mm in-plane, 1.5- to 4-mm section thickness), number of slices
(range, 9–64), acquisition time (range, 2–27 minutes), and imaging
time poststroke (range, 6 hours to 4 weeks). A large high-resolution
highly sampled q-space data set similar to that with connectome
imaging would be ideal to determine the value of higher resolution
and sampling for the use of beyond-DTI methods in stroke.
Different acquisition parameters may lead to different results.
This possibility is best known in DTI; for example, Table 1 in
Barrio-Arranz et al99 describes results from 14 studies of different
spatial resolutions, b-values, or diffusion directions and how FA
and MD changed. Less has been studied with beyond-DTI methods, but for example, NODDI results are a function of b-values
and gradient directions.100 This finding again is motivation for
large rigorous studies in stroke with overcomplete sampling and
outcome measures for validation.
AJNR Am J Neuroradiol 43:661–69 May 2022
www.ajnr.org
665
Analysis methods tended to be from either ROIs in parts or
all of the cortical spinal tract (PLIC, the cerebral peduncle, and
the corona radiata) and/or from manually identified ischemic
regions. Note that it has been shown in DTI that different fitting
methods (linear, nonlinear) give significantly different results.101
Beyond-DTI methods such as a more rapid NODDI fitting
method102 also are known to affect results compared with the
original slower NODDI fitting method. A few studies used tractbased analyses and/or voxel-based analyses. Atlas-based regions
were often used, though it is not known how well the atlas
matches tracts in stroke regions with disrupted fiber tracts.
Trade-offs with SNR and with how different beyond-DTI parameters vary with voxel size are yet to be studied in subjects with
stroke. Even if ROI analysis methods are used, it could be that
imaging with smaller voxels could better inform microstructure
and tractography methods.
Note that for predicting stroke recovery and following stroke
recovery, we are interested not only in local stroke effects but also
global stroke effects. The best focus and most relevant information to understand and predict adaptations are still not known.
For example, some have found the PLIC to be the most predictive
of future motor function, even if the stroke was not in the PLIC.
This finding may be because the PLIC is a hub with incoming
sensory neurons and outgoing motor control neurons and the
way that stroke, even in other regions, affects the PLIC is telling.
Alternatively, this may simply be from the small sample size and
limited comparisons with other fiber tracts; other regions or composites of regions may provide improved predictive information.
The different acquisitions, nonstandard processing, and few
studies with relatively few subjects that include motor stroke outcomes mean that larger studies with FM scores and current acquisition techniques including simultaneous multislice methods103 are
needed to better understand the impact of beyond-DTI methods.
While the scan times vary widely, from 2 to 27 minutes (Online
Supplemental Data), this variation is, in part, due to limited coverage, spatial resolution, and q-space sampling choices. Acquisition
requirements for optimal performance for the different methods
are not yet known in stroke. It is likely that all of the beyond-DTI
methods can be made clinically practical (,5 minutes) with
advanced acquisition (simultaneous multislice) and deep learning
techniques,104 though those techniques are still evolving and being
evaluated. Direct comparisons between the beyond-DTI methods
and investigation into combining parameters from different methods should also be performed in the context of stroke.
Other Considerations
As pointed out by Kim et al,38 90% of neurologic biomarker studies, including most DTI studies as well as most nonimaging
methods, did not use an independent set of stroke data to crossvalidate their prediction model. Thus, the results are a best case
of predicted correlation, retrospectively, with the given data set.
The beyond-DTI studies reported in the tables here did not perform cross-validation, though 1 small study used k-fold cross-validation96 to have more confidence in the predictive value. The
studies also did not identify or discuss minimally clinically important differences,38 which will be essential to further develop
and translate these promising new methods to clinical use.
666
DiBella
May 2022 www.ajnr.org
Another open question is how outcome predictions vary as a
function of therapy. This question includes acute treatments such
as tPA or thrombectomy and therapies such as brain-computer
interface methods. The brain-computer interface has shown
promising results in small studies of people with chronic stroke;
recovery may be enhanced, especially if such therapies are started
early after stroke. Indeed, advanced diffusion-based metrics may
be able to predict who would benefit the most from the braincomputer interface or other types of therapies; plasticity or reorganization changes may be measurable with beyond-DTI parameters while confounding the interpretation of FA. The ability of
diffusion methods to differentiate ipsilateral-versus-contralateral
contributions of motor tracts to upper extremity function may
also inform stroke rehabilitation approaches.105
The many exciting new advances in neuroimaging and particularly in beyond-DTI methods are likely to have a highly significant impact in the context of predicting stroke recovery, for
improving our understanding of brain changes after stroke, and
for providing unique advantages when selecting personalized
stroke therapy. However, this review also makes clear that there
is a need for rigorous studies to better evaluate and translate these
microstructural mapping methods to clinical application.
Disclosure forms provided by the authors are available with the full text and
PDF of this article at www.ajnr.org.
REFERENCES
1. Roger VL, Go AS, Lloyd-Jones DM, et al; American Heart Association
Statistics Committee and Stroke Statistics Subcommittee. Heart disease and stroke statistics–2011 update: a report from the American
Heart Association. Circulation 2011;123:e18–e209 CrossRef Medline
2. Kumar P, Kathuria P, Nair P, et al. Prediction of upper limb motor
recovery after subacute ischemic stroke using diffusion tensor
imaging: a systematic review and meta-analysis. J Stroke 2016;18:50–
59 CrossRef Medline
3. Puig J, Blasco G, Schlaug G, et al. Diffusion tensor imaging as a
prognostic biomarker for motor recovery and rehabilitation after
stroke. Neuroradiology 2017;59:343–51 CrossRef Medline
4. Moura LM, Luccas R, de Paiva JP, et al. Diffusion tensor imaging
biomarkers to predict motor outcomes in stroke: a narrative
review. Front Neurol 2019;10:445 Medline
5. Hui ES, Fieremans E, Jensen JH, et al. Stroke assessment with diffusional kurtosis imaging. Stroke 2012;43:2968–73 CrossRef Medline
6. Granziera C, Daducci A, Meskaldji DE, et al. A new early and automated MRI-based predictor of motor improvement after stroke.
Neurology 2012;79:39–46 CrossRef Medline
7. Mastropietro A, Rizzo G, Fontana L, et al. Microstructural characterization of corticospinal tract in subacute and chronic stroke
patients with distal lesions by means of advanced diffusion MRI.
Neuroradiology 2019;61:1033–45 CrossRef Medline
8. Cramer SC, Nelles G, Benson RR, et al. Functional MRI study of
subjects recovered from hemiparetic stroke. Stroke 1997;28:2518–
27 CrossRef Medline
9. Hatem SM, Saussez G, Della Faille M, et al. Rehabilitation of motor
function after stroke: a multiple systematic review focused on
techniques to stimulate upper extremity recovery. Front Hum
Neurosci 2016;10:442 CrossRef Medline
10. Liang D, Bhatta S, Gerzanich V, et al. Cytotoxic edema: mechanisms of pathological cell swelling. Neurosurg Focus 2007;22:E2
CrossRef Medline
11. Goyal M, Ospel JM, Menon B, et al. Challenging the ischemic core
concept in acute ischemic stroke imaging. Stroke 2020;51:3147–55
CrossRef Medline
12. Hori N, Carpenter DO. Functional and morphological changes
induced by transient in vivo ischemia. Exp Neurol 1994;129:279–
89 CrossRef Medline
13. Baron CA, Kate M, Gioia L, et al. Reduction of diffusion-weighted
imaging contrast of acute ischemic stroke at short diffusion
times. Stroke 2015;46:2136–41 CrossRef Medline
14. Zuo M, Guo H, Wan T, et al. Wallerian degeneration in experimental focal cortical ischemia. Brain Res Bull 2019;149:194–202
CrossRef Medline
15. Jones TA, Schallert T. Overgrowth and pruning of dendrites in
adult rats recovering from neocortical damage. Brain Res
1992;581:156–60 CrossRef Medline
16. Jones TA, Adkins DL. Motor system reorganization after stroke:
stimulating and training toward perfection. Physiology (Bethesda)
2015;30:358–70 CrossRef Medline
17. Alexander DC, Dyrby TB, Nilsson M, et al. Imaging brain microstructure with diffusion MRI: practicality and applications. NMR
Biomed 2019;32:e3841 CrossRef Medline
18. Kiselev VG. Fundamentals of diffusion MRI physics. NMR
Biomed 2017;30 CrossRef Medline
19. Hagmann P, Jonasson L, Maeder P, et al. Understanding diffusion
MR imaging techniques: from scalar diffusion-weighted imaging
to diffusion tensor imaging and beyond. Radiographics 2006;26
(Suppl 1):S205–23 Medline
20. Novikov DS, Kiselev VG, Jespersen SN. On modeling. Magn Reson
Med 2018;79:3172–93 CrossRef Medline
21. Tuch DS. Q-ball imaging. Magn Reson Med 2004;52:1358–72
CrossRef Medline
22. Concha L. A macroscopic view of microstructure: using diffusionweighted images to infer damage, repair, and plasticity of white
matter. Neuroscience 2014;276:14–28 CrossRef Medline
23. Fukutomi H, Glasser MF, Zhang H, et al. Neurite imaging reveals
microstructural variations in human cerebral cortical gray matter. Neuroimage 2018;182:488–99 CrossRef Medline
24. Palombo M, Ianus A, Guerreri M, et al. SANDI: A compartmentbased model for non-invasive apparent soma and neurite imaging
by diffusion MRI. Neuroimage 2020;215:116835 CrossRef Medline
25. Boscolo Galazzo I, Brusini L, Obertino S, et al. On the viability of
diffusion MRI-based microstructural biomarkers in ischemic
stroke. Front Neurosci 2018;12:92 CrossRef Medline
26. Schaefer PW, Grant PE, Gonzalez RG. Diffusion-weighted MR
imaging of the brain. Radiology 2000;217:331–45 CrossRef Medline
27. Duong TQ, Ackerman JJ, Ying HS, et al. Evaluation of extra- and
intracellular apparent diffusion in normal and globally ischemic
rat brain via 19F NMR. Magn Reson Med 1998;40:1–13 CrossRef
Medline
28. Jelescu IO, Veraart J, Fieremans E, et al. Degeneracy in model parameter estimation for multi-compartmental diffusion in neuronal tissue. NMR Biomed 2016;29:33–47 CrossRef Medline
29. Dhital B, Reisert M, Kellner E, et al. Intra-axonal diffusivity in brain
white matter. Neuroimage 2019;189:543–50 CrossRef Medline
30. Lampinen B, Szczepankiewicz F, Martensson J, et al. Towards
unconstrained compartment modeling in white matter using diffusion-relaxation MRI with tensor-valued diffusion encoding.
Magn Reson Med 2020;84:1605–23 CrossRef Medline
31. Budde MD, Frank JA. Neurite beading is sufficient to decrease the
apparent diffusion coefficient after ischemic stroke. Proc Natl
Acad Sci U S A 2010;107:14472–77 CrossRef Medline
32. Datar A, Ameeramja J, Bhat A, et al. The roles of microtubules and
membrane tension in axonal beading, retraction, and atrophy.
Biophys J 2019;117:880–91 CrossRef Medline
33. Brugieres P, Thomas P, Maraval A, et al. Water diffusion compartmentation at high b values in ischemic human brain. AJNR Am J
Neuroradiol 2004;25:692–68 Medline
34. Nilsson M, van Westen D, Stahlberg F, et al. The role of tissue
microstructure and water exchange in biophysical modelling of
diffusion in white matter. MAGMA 2013;26:345–70 CrossRef
Medline
35. Lampinen B, Latt J, Wasselius J, et al. Time dependence in diffusion MRI predicts tissue outcome in ischemic stroke patients.
Magn Reson Med 2021;86:754–64 CrossRef Medline
36. Kruetzelmann A, Kohrmann M, Sobesky J, et al. Pretreatment diffusion-weighted imaging lesion volume predicts favorable outcome
after intravenous thrombolysis with tissue-type plasminogen activator in acute ischemic stroke. Stroke 2011;42:1251–54 CrossRef
Medline
37. Campbell BC, Tu HT, Christensen S, et al. Assessing response to
stroke thrombolysis: validation of 24-hour multimodal magnetic
resonance imaging. Arch Neurol 2011;69:45–50 CrossRef Medline
38. Kim B, Winstein C. Can neurological biomarkers of brain impairment be used to predict poststroke motor recovery? A systematic
review. Neurorehabil Neural Repair 2017;31:3–24 CrossRef Medline
39. Kwah LK, Herbert RD. Prediction of walking and arm recovery after stroke: a critical review. Brain Sci 2016;6:53 CrossRef Medline
40. Hemmen TM, Ernstrom K, Raman R. Two-hour improvement of
patients in the National Institute of Neurological Disorders and
Stroke trials and prediction of final outcome. Stroke 2011;42:
3163–67 CrossRef Medline
41. Prabhakaran S, Zarahn E, Riley C, et al. Inter-individual variability
in the capacity for motor recovery after ischemic stroke. Neurorehabil Neural Repair 2008;22:64–71 CrossRef Medline
42. Krakauer JW, Marshall RS. The proportional recovery rule for
stroke revisited. Ann Neurol 2015;78:845–47 CrossRef Medline
43. Winters C, van Wegen EE, Daffertshofer A, et al. Generalizability
of the proportional recovery model for the upper extremity after
an ischemic stroke. Neurorehabil Neural Repair 2015;29:614–22
CrossRef Medline
44. Stinear CM, Byblow WD, Ackerley SJ, et al. Proportional motor recovery after stroke: implications for trial design. Stroke 2017;
48:795–98 CrossRef Medline
45. Fugl-Meyer AR, Jaasko L, Leyman I, et al. The post-stroke hemiplegic patient, 1; a method for evaluation of physical performance.
Scand J Rehabil Med 1975;7:13–31 Medline
46. Hawe RL, Scott SH, Dukelow SP. Taking proportional out of
stroke recovery. Stroke 2018 Dec 7. [Epub ahead of print] CrossRef
Medline
47. Hope TM, Friston K, Price CJ, et al. Recovery after stroke: not so
proportional after all? Brain 2019;142:15–22 CrossRef Medline
48. Farr TD, Wegener S. Use of magnetic resonance imaging to predict outcome after stroke: a review of experimental and clinical
evidence. J Cereb Blood Flow Metab 2010;30:703–17 CrossRef
Medline
49. Sun C, Liu X, Bao C, et al. Advanced non-invasive MRI of neuroplasticity in ischemic stroke: techniques and applications. Life Sci
2020;261:118365 CrossRef Medline
50. Albers GW, Thijs VN, Wechsler L, et al; DEFUSE Investigators.
Magnetic resonance imaging profiles predict clinical response to
early reperfusion: the diffusion and perfusion imaging evaluation
for understanding stroke evolution (DEFUSE) study. Ann Neurol
2006;60:508–17 CrossRef Medline
51. Qiu M, Darling WG, Morecraft RJ, et al. White matter integrity is
a stronger predictor of motor function than BOLD response in
patients with stroke. Neurorehabil Neural Repair 2011;25:275–84
CrossRef Medline
52. Carter AR, Shulman GL, Corbetta M. Why use a connectivitybased approach to study stroke and recovery of function?
NeuroImage 2012;62:2271–80 CrossRef Medline
53. Zhu LL, Lindenberg R, Alexander MP, et al. Lesion load of the corticospinal tract predicts motor impairment in chronic stroke.
Stroke 2010;41:910–15 CrossRef Medline
54. Kumar MA, Vangala H, Tong DC, et al. MRI guides diagnostic
approach for ischaemic stroke. J Neurol Neurosurg Psychiatry
2011:83;1201–05 CrossRef Medline
55. Song J, Nair VA, Young BM, et al. DTI measures track and predict
motor function outcomes in stroke rehabilitation utilizing BCI
technology. Front Hum Neurosci 2015;9:195 CrossRef Medline
AJNR Am J Neuroradiol 43:661–69 May 2022
www.ajnr.org
667
56. Puig J, Pedraza S, Blasco G, et al. Wallerian degeneration in the
corticospinal tract evaluated by diffusion tensor imaging correlates with motor deficit 30 days after middle cerebral artery ischemic stroke. AJNR Am J Neuroradiol 2010;31:1324–30 CrossRef
Medline
57. Groisser BN, Copen WA, Singhal AB, et al. Corticospinal tract diffusion abnormalities early after stroke predict motor outcome.
Neurorehabil Neural Repair 2014;28:751–60 CrossRef Medline
58. Jang SH, Lee J, Lee MY, et al. Prediction of motor outcome using
remaining corticospinal tract in patients with pontine infarct: diffusion tensor imaging study. Somatosens Mot Res 2016;33:99–103
CrossRef Medline
59. Pannek K, Chalk JB, Finnigan S, et al. Dynamic corticospinal white
matter connectivity changes during stroke recovery: a diffusion
tensor probabilistic tractography study. J Magn Reson Imaging
2009;29:529–36 CrossRef Medline
60. Kim KH, Kim YH, Kim MS, et al. Prediction of motor recovery
using diffusion tensor tractography in supratentorial stroke
patients with severe motor involvement. Ann Rehabil Med
2015;39:570–76 CrossRef Medline
61. Frank LR. Anisotropy in high angular resolution diffusionweighted MRI. Magn Reson Med 2001;45:935–39 CrossRef Medline
62. Tuch DS, Reese TG, Wiegell MR, et al. High angular resolution diffusion imaging reveals intravoxel white matter fiber heterogeneity. Magn Reson Med 2002;48:577–82 CrossRef Medline
63. Wedeen VJ, Wang RP, Schmahmann JD, et al. Spectrum magnetic
resonance imaging (DSI) tractography of crossing fibers.
Neuroimage 2008;41:1267–77 CrossRef Medline
64. Van AT, Granziera C, Bammer R. An introduction to model-independent diffusion magnetic resonance imaging. Top Magn Reson
Imaging 2010;21:339–54 CrossRef Medline
65. Jones DK, Knosche TR, Turner R. White matter integrity, fiber
count, and other fallacies: The do’s and don’ts of diffusion MRI.
Neuroimage 2013;73:239–54 CrossRef Medline
66. Lakhani DA, Schilling KG, Xu J, et al. Advanced multicompartment
diffusion MRI models and their application in multiple sclerosis.
AJNR Am J Neuroradiol 2020;41:751–57 CrossRef Medline
67. Jensen JH, Helpern JA, Ramani A, et al. Diffusional kurtosis imaging: the quantification of non-Gaussian water diffusion by means
of magnetic resonance imaging. Magn Reson Med 2005;53:1432–40
CrossRef Medline
68. Lu H, Jensen JH, Ramani A, et al. Three-dimensional characterization of non-Gaussian water diffusion in humans using diffusion
kurtosis imaging. NMR Biomed 2006;19:236–47 CrossRef Medline
69. Cheung JS, Wang E, Lo EH, et al. Stratification of heterogeneous
diffusion MRI ischemic lesion with kurtosis imaging: evaluation
of mean diffusion and kurtosis MRI mismatch in an animal
model of transient focal ischemia. Stroke 2012;43:2252–54
CrossRef Medline
70. Hui ES, Du F, Huang S, et al. Spatiotemporal dynamics of diffusional kurtosis, mean diffusivity and perfusion changes in experimental stroke. Brain Res 2012;1451:100–09 CrossRef Medline
71. Sun PZ, Wang Y, Mandeville E, et al. Validation of fast diffusion
kurtosis MRI for imaging acute ischemia in a rodent model of
stroke. NMR Biomed 2014;27:1413–18 CrossRef Medline
72. Umesh Rudrapatna S, Wieloch T, Beirup K, et al. Can diffusion
kurtosis imaging improve the sensitivity and specificity of detecting microstructural alterations in brain tissue chronically after
experimental stroke? Comparisons with diffusion tensor imaging
and histology. Neuroimage 2014;97:363–73 CrossRef Medline
73. Weber RA, Hui ES, Jensen JH, et al. Diffusional kurtosis and diffusion tensor imaging reveal different time-sensitive stroke-induced
microstructural changes. Stroke 2015;46:545–50 CrossRef Medline
74. Wu Y, Kim J, Chan ST, et al. Comparison of image sensitivity
between conventional tensor-based and fast diffusion kurtosis
imaging protocols in a rodent model of acute ischemic stroke.
NMR Biomed 2016;29:625–30 CrossRef Medline
668
DiBella
May 2022
www.ajnr.org
75. Zhang S, Yao Y, Shi J, et al. The temporal evolution of diffusional
kurtosis imaging in an experimental middle cerebral artery occlusion (MCAO) model. Magn Reson Imaging 2016;34:889–95
CrossRef Medline
76. Wang E, Wu Y, Cheung JS, et al. pH imaging reveals worsened tissue acidification in diffusion kurtosis lesion than the kurtosis/diffusion lesion mismatch in an animal model of acute stroke. J
Cereb Blood Flow Metab 2017;37:3325–33 CrossRef Medline
77. Lu D, Jiang Y, Ji Y, et al. JOURNAL CLUB: evaluation of diffusion
kurtosis imaging of stroke lesion with hemodynamic and metabolic MRI in a rodent model of acute stroke. AJR Am J Roentgenol
2018;210:720–27 CrossRef Medline
78. Bay V, Kjolby BF, Iversen NK, et al. Stroke infarct volume estimation in fixed tissue: Comparison of diffusion kurtosis imaging to
diffusion weighted imaging and histology in a rodent MCAO
model. PLoS One 2018;13:e0196161 CrossRef Medline
79. Cheung J, Doerr M, Hu R, et al. Refined ischemic penumbra imaging with tissue pH and diffusion kurtosis magnetic resonance
imaging. Transl Stroke Res 2021;12:742–53 CrossRef Medline
80. Puig J, Pedraza S, Blasco G, et al. Acute damage to the posterior
limb of the internal capsule on diffusion tensor tractography as
an early imaging predictor of motor outcome after stroke. AJNR
Am J Neuroradiol 2011;32:857–63 CrossRef Medline
81. Hansen B, Jespersen SN. Recent developments in fast kurtosis
imaging. Frontiers in Physics 2017;5 CrossRef
82. Spampinato MV, Chan C, Jensen JH, et al. Diffusional kurtosis
imaging and motor outcome in acute ischemic stroke. AJNR Am J
Neuroradiol 2017;38:1328–34 CrossRef Medline
83. Li C, Lan C, Zhang X, et al. Evaluation of diffusional kurtosis
imaging in sub-acute ischemic stroke: comparison with rehabilitation treatment effect. Cell Transplant 2019;28:1053–61 CrossRef
Medline
84. Yu X, Jiaerken Y, Wang S, et al. Changes in the corticospinal tract
beyond the ischemic lesion following acute hemispheric stroke:
a diffusion kurtosis imaging study. J Magn Reson Imaging
2020;52:512–19 CrossRef Medline
85. Chung S, Fieremans E, Wang X, et al. White matter tract integrity:
an indicator of axonal pathology after mild traumatic brain
injury. J Neurotrauma 2018;35:1015–20 CrossRef Medline
86. Wedeen VJ, Hagmann P, Tseng WY, et al. Mapping complex tissue
architecture with diffusion spectrum magnetic resonance imaging. Magn Reson Med 2005;54:1377–86 CrossRef Medline
87. Jiang Q, Zhang ZG, Chopp M. MRI evaluation of white matter recovery after brain injury. Stroke 2010;41:S112–13 CrossRef Medline
88. Tang PF, Ko YH, Luo ZA, et al. Tract-specific and region of interest
analysis of corticospinal tract integrity in subcortical ischemic
stroke: reliability and correlation with motor function of affected
lower extremity. AJNR Am J Neuroradiol 2010;31:1023–30 CrossRef
Medline
89. Granziera C, Daducci A, Gigandet X, et al. Diffusion spectrum
imaging after stroke shows structural changes in the contra-lateral motor network correlating with functional recovery. In:
Proceedings of the International Society for Magnetic Resonance in
Medicine Conference, Montreal, Quebec, Canada. May 7–13, 2011
90. Ozarslan E, Koay CG, Shepherd TM, et al. Mean apparent propagator (MAP) MRI: a novel diffusion imaging method for mapping
tissue microstructure. Neuroimage 2013;78:16–32 CrossRef Medline
91. Brusini L, Obertino S, Galazzo IB, et al. Ensemble average propagator-based detection of microstructural alterations after stroke. Int
J Comput Assist Radiology Surg 2016;11:1585–97 CrossRef Medline
92. Zhang H, Schneider T, Wheeler-Kingshott CA, et al. NODDI: practical in vivo neurite orientation dispersion and density imaging of
the human brain. Neuroimage 2012;61:1000–16 CrossRef Medline
93. Anderova M, Vorisek I, Pivonkova H, et al. Cell death/proliferation
and alterations in glial morphology contribute to changes in diffusivity in the rat hippocampus after hypoxia-ischemia. J Cereb
Blood Flow Metab 2011;31:894–907 CrossRef Medline
94. Bagdasarian FA, Yuan X, Athey J, et al. NODDI highlights recovery
mechanisms in white and gray matter in ischemic stroke following human stem cell treatment. Magn Reson Med 2021;86:3211–23
CrossRef Medline
95. Adluru G, Gur Y, Anderson J, et al. Assessment of white matter
microstructure in stroke patients using NODDI. Annu Int Conf
IEEE Eng Med Biol Soc 2014;2014:742–45 CrossRef Medline
96. Hodgson K, Adluru G, Richards LG, et al. Predicting motor outcomes in stroke patients using diffusion spectrum MRI microstructural measures. Front Neurol 2019;10:72 CrossRef Medline
97. Wang Z, Zhang S, Liu C, et al. A study of neurite orientation dispersion and density imaging in ischemic stroke. Magn Reson
Imaging 2019;57:28–33 CrossRef Medline
98. Latt J, Nilsson M, van Westen D, et al. Diffusion-weighted MRI
measurements on stroke patients reveal water-exchange mechanisms in sub-acute ischaemic lesions. NMR Biomed 2009;22:619–
28 CrossRef Medline
99. Barrio-Arranz G, de Luis GR, Tristan-Vega A, et al. Impact of MR
acquisition parameters on DTI scalar indexes: a tractography
based approach. PLoS One 2015;10:e0137905 CrossRef Medline
100. Parvathaneni P, Nath V, Blaber JA, et al. Empirical reproducibility,
sensitivity, and optimization of acquisition protocol, for neurite
101.
102.
103.
104.
105.
orientation dispersion and density imaging using AMICO. Magn
Reson Imaging 2018;50:96–109 CrossRef Medline
Bergamino M, Keeling EG, Walsh RR, et al. Systematic assessment
of the impact of DTI methodology on fractional anisotropy measures in Alzheimer’s disease. Tomography 2021;7:20–38 CrossRef
Medline
Daducci A, Canales-Rodriguez EJ, Zhang H, et al. Accelerated
Microstructure Imaging via Convex Optimization (AMICO)
from diffusion MRI data. Neuroimage 2015;105:32–44 CrossRef
Medline
HashemizadehKolowri K, Chen RR, Adluru G, et al. Simultaneous
multi-slice image reconstruction using regularized image domain
split slice-GRAPPA for diffusion MRI. Med Image Anal 2021;70:
102000 CrossRef Medline
Ye C, Li Y, Zeng X. An improved deep network for tissue microstructure estimation with uncertainty quantification. Med Image
Anal 2020;61:101650 CrossRef Medline
Bundy DT, Souders L, Baranyai K, et al. Contralesional brain-computer interface control of a powered exoskeleton for motor recovery in chronic stroke survivors. Stroke 2017;48:1908–15 CrossRef
Medline
AJNR Am J Neuroradiol 43:661–69 May 2022
www.ajnr.org
669