Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2019 Jul 1.
Published in final edited form as: J Neurosci Methods. 2018 Apr 21;304:66–75. doi: 10.1016/j.jneumeth.2018.04.010

A Comparison of Seven Different DTI-derived Estimates of Corticospinal Tract Structural Characteristics in Chronic Stroke Survivors

Bokkyu Kim 1, Beth E Fisher 2,3, Nicolas Schweighofer 2,4, Richard M Leahy 5,6, Justin P Haldar 5,6, Soyoung Choi 4, Dorsa B Kay 4, James Gordon 2, Carolee J Winstein 2,3
PMCID: PMC5984168  NIHMSID: NIHMS963800  PMID: 29684462

Abstract

Background

Different diffusion tensor imaging (DTI) has been used to estimate corticospinal tract (CST) structure in the context of stroke rehabilitation research. However, there is no gold standard for the estimate of CST structure in chronic stroke survivors. This study aims to determine the most accurate DTI-derived CST estimate that is associated with a clinical motor outcome measure.

Methods

We obtained imaging and behavioral data from a phase-I stroke rehabilitation clinical trial. We included thirty seven chronic stroke survivors with mild-to-moderate motor impairment. Imaging data were processed using BrainSuite16a software. We calculated mean FA for each of 7 different ROIs/VOIs that include manually drawn 2-D ROIs and 3-D VOIs of CST from individual tractography or standard atlas. We compared ipsi- and contralesional CST FA for each method. Partial correlation was conducted between each CST FA asymmetry index and a time-based motor outcome measure, controlling for age and chronicity.

Results

Ipsilesional CST FA was significantly lower than contralesional CST FA for each of the 7 methods. Only CST FA asymmetry from the 3-D individual CST tractography showed a significant correlation with the primary motor outcome (r=0.46, p=0.005), while CST FA from the other six methods did not.

Comparison with Existing Methods

Compared to the six other methods, CST FA asymmetry from 3-D individual tractography is the most accurate estimate of CST structure in this cohort of stroke survivors.

Conclusion

We recommend this method for future research seeking to understand brain-behavior mechanisms of motor recovery in chronic stroke survivors.

Keywords: Stroke rehabilitation, Diffusion Tensor Imaging, Corticospinal Tract

1. Introduction

The corticospinal tract (CST) delivers motor commands from primary motor and sensory cortices to the spinal cord through direct (lateral and medial corticospinal tract) and indirect pathways. As such, the amount of post-stroke damage affecting this descending motor pathway is a crucial factor that determines the level of motor impairment following stroke (Corbetta et al., 2015; Jang, 2011). Diffusion tensor imaging (DTI)-derived estimate of CST structural damage to the CST has been used to predict motor recovery in stroke survivors. DTI-derived metrics of CST (Koyama et al., 2013; Lindenberg et al., 2012, 2010, Puig et al., 2013, 2011), specifically fractional anisotropy (FA) asymmetry and the ratio between ipsi- and contra-lesional CST, are the most frequently used predictor variables in prognostic studies (Lindenberg et al., 2010; Puig et al., 2013; Stinear et al., 2012). In general, FA of ipsilesional CST is decreased after stroke, leading to an increase in FA asymmetry (Stinear et al., 2012).

Although DTI-derived estimates of ipsilesional CST are widely used, the specific estimation method varies considerably (Lindenberg et al., 2010); as such, comparison across studies is challenging. The four most common regions/volumes of interest that investigators have used to estimate CST structure are: 1) a 2-dimensional region of interest (2-D ROI) at the posterior limb of the internal capsule (PLIC); 2) a 2-D ROI at the cerebral peduncle (CP); 3) a 3-dimensional (3-D) volume of interest (3-D VOI) of CST from individual stroke patient’s CST tractography; and 4) a 3-D VOI of CST from non-disabled adults’ CST template (Kim and Winstein, 2017).

For 2-D ROI-based methods, a PLIC or CP ROI is manually drawn on structural images (Koyama et al., 2012). To perform CST structural estimation using 3-D CST VOI, one can reconstruct the individual CST of each individual using diffusion tensor-based tractography (Lindenberg et al., 2012, 2010; Yu et al., 2009). Template CST tractography is reconstructed from age-matched non-disabled participants’ DTI data, or an established template standard CST atlas can be used (Park et al., 2013).

There are advantages and disadvantages of each estimate of CST structure. The benefits of using 2-D PLIC or CP ROIs are that both ROIs are easily identified on structural images, and the FA value of these regions has been shown to provide an accurate estimation of CST structural damage (Kuzu et al., 2012). A decreased FA at these regions is thought to reflect the Wallerian degeneration that occurs across the entire CST after stroke (Koyama et al., 2012; Kuzu et al., 2012; Lindenberg et al., 2012). Thus, determination of the DTI-derived metrics at a remote CST section, such as PLIC or CP, is assumed to represent the degree of degeneration of the entire CST without underestimation. However, the ROI-based methods may be biased, particularly when PLIC or CP ROIs are manually drawn (Lindenberg et al., 2010). To control for possible bias introduced by these manual methods, some researchers have instead used PLIC or CP sections of CST tractography or established subcortical white matter atlases (Park et al., 2013).

Compared with 2-D ROI-based methods, individual tractography-based CST estimation methods can represent the entire CST structure using the subject’s own reconstructed fiber tracts. However, the individual CST method may not be appropriate for cases in which some if not most CST fibers cannot be traced using diffusion tensor-based tractography (Cho et al., 2007a, 2007b).

In general, template-based CST estimates are considered a more objective method compared to ROI-based methods, as it relies on automated processes known to reduce operator-dependent bias (Park et al. 2003). However, the brain of chronic stroke survivors may present with an aberrant ipsilesional CST trajectory due in part to significant subcortical white matter atrophy; in this case, an estimate of CST structure that relies on template CST volume will likely be incorrect (Jang, 2011).

Although these different DTI-derived estimates for CST structure are widely used in stroke rehabilitation research, there is no gold standard which most accurately represents CST structural characteristics in chronic stroke survivors. Thus, this study aims 1) to determine which method most accurately estimates CST structural damage, and 2) to examine the degree to which the measure of CST structural damage correlates with a well-known clinical motor outcome in chronic stroke survivors with mild-to-moderate motor impairment.

To determine the most accurate estimate of CST structure, each method was evaluated based on three criteria: 1) the method can capture a significant decrease in ipsilesional CST fractional anisotropy (FA) compared to the FA of contralesional CST; 2) CST FA asymmetry range falls within the normative range between −0.03 and 0.25; and 3) significant relationship between CST FA asymmetry and clinical motor outcome.

Our expected findings are that the most accurate method for this cohort will be the 3-D individual tractography-based CST method, and the CST FA asymmetry derived from this method will result in the strongest correlation with our primary motor outcome. Our hypotheses are based on two primary assumptions: 2-D ROIs may not represent the entire CST structure, and 3-D template CST VOI may not represent the distorted ipsilesional CST trajectory commonly seen in the chronic stroke brain.

2. Methods

2.1. Participants

The clinical and neuroimaging data were from a single-site randomized trial of stroke rehabilitation conducted in the Motor Behavior and Neurorehabilitation Laboratory at Division of Biokinesiology and Physical Therapy, University of Southern California (ClinicalTrials.gov ID: NCT 01749358). The clinical trial was conducted in accordance with the Delaration of Helsinki and all procedures were carried out with the adequate understanding and written consent of the participants. The clinical trial was approved by the Institutional Review Board of the University of Southern California. The purpose of this clinical trial was to determine the effect of different doses of therapy on motor outcomes in chronic stroke. Predefined inclusion and exclusion criteria for the RCT are described in Supplementary Table 1. For this project, a total of 37 out of 42 trial participants’ data met inclusion criteria that required a complete set of baseline clinical motor outcome scores and DTI neuroimaging. Five data sets were excluded because of missing neuroimaging or artifacts present on imaging.

2.2. Clinical motor outcome measure

We utilized an arm-specific clinical motor outcome measure. Specifically, we employed a subset of laboratory-based Wolf Motor Function Test (WMFT). WMFT time score is a reliable and valid method to evaluate UE motor performance after stroke, particularly in the research environment (Lin et al., 2009; Wolf et al., 2001). This measure includes fifteen timed motor tasks. The timed tasks can be distributed into two task categories: 1) tasks related to joint-segment movements, and 2) tasks related to integrative functional movements (Wolf et al., 2001). The joint-segment movement tasks are primarily those with proximal joint control (e.g., shoulder and elbow), while the integrative functional tasks require some level of hand dexterity for object manipulation. Given our sample of those with mild-to-moderate motor impairment, tasks related to proximal joint-segment movements had ceiling effects on assessing the upper extremity motor performance (Appendix 1). Thus, we employed the WMFT-distal time score (WMFT-distal), the log-transformed average time score for the integrative functional task items (i.e., distal arm control task items), to minimize the ceiling effect from the tasks associated with joint-segment movements.

2.3. MRI acquisition

We used a 3 Tesla GE Signa Excite MRI scanner to acquire high resolution T1-weighted and diffusion-weighted images with the following acquisition parameters:

2.3.1. T1-weighted image

A set of coronal scout scans covering the entire brain was acquired first to define the field-of-view, and then 124 sagittal slices covering the entire brain were acquired. Next a sagittal anatomic 3-D volumetric study designed to increase spatial resolution and tissue contrast was acquired using a gradient-echo (SPGR) T-1 weighted series with TR=24, TE=3.5ms, flip angle=20°, field of view (FoV)=24 cm, and slice thickness=1.2 mm with no gaps. This procedure was completed in approximately 10 minutes.

2.3.2. Diffusion MRI

The diffusion MRI scan used a single shot spin echo EPI pulse sequence using the following parameters: TR=10,000 ms, TE=88 ms, 75 axial slices, FOV=256 mm, slice thickness=2.0 mm, Matrix = 128 × 128, b-value=1000, 64 diffusion gradient directions. The diffusion MRI sequence was completed in approximately 10 minutes.

2.4 MRI data analysis

The raw DICOM files were converted to NIFTI format using the MRIcron DICOM to NIFTI function (http://people.cas.sc.edu/rorden/mricron/dcm2nii.html). BrainSuite software (http://brainsuite.org) was used to process T1-weighted images and DTI. The MRI process sequence for T1-weighted images includes cortical surface extraction and surface-volume registration (SVREG). The cortical surface extraction includes skull striping (Sandor and Leahy, 1997), intensity non-uniformity correction (Shattuck et al., 2001), tissue classification (Shattuck et al., 2001), inner cortex masking, surface generation, and hemisphere labeling. This process was automatically performed with pre-determined parameters. A quality check was performed at each step, and if the result of the step was incorrect due to presence of the stroke lesion, manual correction was performed. Specifically, results from the tissue classification and inner cortex masking were frequently incorrect due to the presence of the lesion. To improve the tissue classification, intensity non-uniformity correction was re-performed using manual bias field correction software and/or Brainsuite’s bias field correction function in iterative mode. In addition, manual mask correction was conducted if the inner cortex mask did not accurately identify the gray/white boundary. This manual correction procedure improved the result of surface-volume registration in Brainsuite. For quality control of the cortical extraction process, images were visually checked by the first author (BK) at each step.

Surface-volume registration (the SVReg function in Brainsuite) is an automated process for co-registration of the T1-weighted image to an atlas brain in which a nonrigid alignment between the subject and atlas is performed, constrained by mapping of the surface of the cerebral cortex of the atlas onto that of the subject (Joshi et al., 2012). We used BrainSuiteAtlas1 (http://brainsuite.org/atlases/), a labeled version of the Colin27 Average Brain (Holmes et al., 1998). The result of applying SVReg to each volume is to automatically label subcortical and cortical structures in the subject space. A visual quality check of the anatomical brain label outputs was performed to assess accuracy of the registration.

For the diffusion MRI preprocessing, we used FSL’s eddy tool (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/eddy) to correct eddy currents and movements in diffusion MRI data (Andersson and Sotiropoulos, 2016). Then, the BrainSuite diffusion pipeline (BDP) was used to process diffusion MRI data. We co-registered the diffusion MRI data to the anatomical T1-weighted images using the INVERSION method (Bhushan et al., 2014); this method includes correction for susceptibility-induced distortions using non-rigid registration. Diffusion tensors were estimated using a weighted linear least squares method, and we computed the scalar diffusion parameters: fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD), and axial diffusivity (AD) based on an eigen-decomposition of the tensors, as described by Kim et al., 2009. After processing of the diffusion MRI data, all images were visually checked for quality control.

After the BDP process, deterministic whole brain diffusion tensor-based tractography was performed using the BrainSuite Diffusion Toolbox. Then we identified the corticospinal tracts (CST) tractography for each hemisphere. CST was identified as streamlines passing through cerebral peduncle (CP), pons on the same side, and originating from primary motor cortex (M1), primary sensory cortex (S1), and supplementary-motor cortex (SMA). Any fibers passing through the corpus callosum, midline of the pons, and cerebellum were excluded. We visually inspected each CST tractography for accuracy (Figure 1). In detail, we applied three steps of filtering to the CST tractography: 1) filter out streamlines to contralateral hemisphere and to cerebellum; 2) select streamlines passing through CP and pons in the same side; 3) select streamlines passing through precentral gyrus (M1), postcentral gyrus (S1), and paracentral lobule (SMA) labels in the same side. If the CST tractography included non-CST streamlines, we manually removed the non-CST streamlines using ‘TRACK FILTERING’ function in the BrainSuite graphic user interface. Briefly, we selected a portion of the non-CST streamlines, and generated a spherical volume of interest in that portion. Streamlines passing through the spherical volume of interest were then excluded (Figure 1-C). If there were multiple non-CST streamlines, we repeated track filtering as necessary.

Figure 1.

Figure 1

Open in a new tab

Visualization of a reconstructed CST tractography from a single participant. Red: right ↔ left, Green: anterior ↔ posterior, Blue: caudal ↔ rostral. The color labels on the superior axial slice indicate brain cortical regions (aqua – left paracentral lobule; purple – left pre-central gyrus; corn flower blue – left post-central gyrus). (A) A representative illustration of CST tractography in the contralesional hemisphere of a chronic stroke survivor with two axial slices showing the sensorimotor cortices and cerebral peduncle regions of interest. (B) An example of inaccurate CST tractography in the lesioned hemisphere. a, b, and c indicate non-CST fibers that need to be excluded manually to improve the accuracy of CST tractography. (C) Manually corrected CST tractography of Figure 1-B. Three spherical volumes of interest (a’, b’, and c’) were used to exclude fibers passing through these volumes.

2.5. DTI-derived estimates of CST structural characteristics

We employed five 3-D VOIs and two 2-D ROIs to determine the most accurate estimate of CST structural characteristics. The mean FA for each 2-D ROI or 3-D VOI was calculated.

2.5.1. Estimate 1 (E1): 3-dimensional (3-D) individual CST tractography

Using the CST tractography of each participant, 3-D tractography-based CST VOI was used to calculate mean FA for the entire CST of each side.

2.5.2. Estimate 2 (E2): 3-D template CST from standard white matter atlas

Using a 3-D template CST from natbrainlab (http://www.natbrainlab.co.uk/atlas-maps) (Thiebaut de Schotten et al., 2011), the mean FA for template CST volume of interest (VOI) was calculated for each side.

2.5.3. Estimate 3 (E3): 3-D template CST from participants’ contralesional CST tractography

To control for the age factor affecting CST trajectory, we also generated a template CST from the contralesional CST tractography of each participant. The CST template from the established white matter atlas was from non-disabled young adults, and therefore may not be appropriate for our stroke cohort. For the left CST template, we used left CST tractography from participants with right hemisphere damage, and vice versa. We determined each CST template using a group-level overlap of each side CST binary masks at a threshold of 50% of the number of participants (Park et al., 2013).

2.5.4. Estimates 4 & 5 (E4 & E5): 3-D posterior limb of the internal capsule (PLIC) and cerebral peduncle (CP) VOIs from standard white matter atlas

To eliminate operator-dependent bias of manual VOI delineation, we computed the mean FA for both 3-D PLIC VOI (E4) and CP VOI (E5) on each side from the Johns Hopkins University (JHU) DTI-based white matter atlas (http://neurovault.org/collections/264/). The white matter atlas was co-registered to the BrainSuite atlas space, and transformed to each participant’s space using the 3-D deformation field generated from SVReg’s mapping between the atlas and the individual T1-weighted image.

2.5.5. Estimates 6 & 7 (E6 & E7): 2-dimensional (2-D) manually drawn PLIC and CP ROIs

2-D PLIC ROI (E6) and CP ROI (E7) on each side were manually drawn on an axial slice of standard T1-weighted BrainSuite atlas image (72nd axial slice [MNI Z coordinate: 0] for PLIC and 54th axial slice [MNI Z coordinate: -18] for CP among 181 axial slices) using BrainSuite software Label tool. In detail, a standard SVReg label file was open on the BrainSuite standard T1-weighted image. For PLIC, the white matter label was selected to make a subcortical white matter mask, then both PLIC labels were manually drawn inside of the white matter mask area. For CP, the brainstem label was selected to make a brainstem mask, then both CP labels were manually drawn inside of the brainstem mask area. Then the ROIs were transformed to each participant’s space using the 3-D deformation field generated from SVReg’s mapping between the atlas and the T1-weighted image. The mean FA for each ROI on each side was calculated.

In order to make sure that 3-D and 2-D template CST labels are symmetrical between hemispheres, we compared the number of voxels of each pair of labels on the atlas space (Table 2). The right CST template VOI from atlas was about 26% bigger than the left CST template VOI. Further, the right subject-based CST template was also about 25% bigger than the left CST template VOI.

Table 2.

Volumes and sizes of template CST labels.

Template Methods Right
[# of voxels]		 Left
[# of voxels]
3-D CST Template from an Atlas 6,903 5,455
3-D CST Template from Subjects 5,842 4,658
3-D PLIC 3,754 3,752
3-D CP 2,278 2,278
2-D PLIC 324 317
2-D CP 262 242
Open in a new tab

2.6. Statistical analysis

Separate paired t-tests were used to test for differences between ipsilesional and contralesional CST FA, with the ipsilesional CST FA expected to be smaller than contralesional CST FA, for each method. In order to test if the FA of ipsilesional CST is decreased in the CST area remote from the lesion, we compared the CST FA between ipsilesional and contralesional sides for each axial slice. This analysis was done for each of the 3-D CST analysis methods, including individual CST tractography, template CST from an atlas, and template CST from the participants’ contralesional CST tractography. For the slice-by-slice comparison, the FA images and CST masks were transformed to BrainSuite atlas standard space. Following this transformation, mean FA for each axial slice for each side was then calculated. Separate paired t-tests were used to test for differences in the FA between ipsilesional and contralesional sides for each axial slice.

To test if CST FA asymmetry is different among different methods, a one-way ANOVA was used with method being the repeated measure. We computed FA asymmetry index [(contralesional CST FA – ipsilesional CST FA)/(contralesional CST FA + ipsilesional CST FA)] in order to control for inter-individual variability in FA. Further, we examined if the range of CST FA asymmetry from each method is within the normative range for this cohort of stroke survivors (between −0.03 and 0.25). An estimate of the normative range of CST FA asymmetry was based on previous studies that reported the CST FA values for a similar stroke population as well as age-matched non-disabled controls (Lindenberg et al., 2012; Puig et al., 2013; Stinear et al., 2007)

Lastly, a partial correlation analyses was used to determine which method quantified the CST that resulted in the strongest relationship with our primary motor outcome score—the WMFT-distal time score. Partial correlation analyses were used to control for age and stroke chronicity.

Bonferroni correction was used to account for the multiple t-tests, the ANOVA post-hoc multiple comparison test, and the multiple correlation analyses. The corrected significance level was set at 0.0071 (alpha = 0.05/7). For the slice by slice analysis for the 3-D CST methods, the corrected significance level was set at 0.000276 (alpha = 0.05/181) as we had 181 axial slices.

3. Results

3.1. Characteristics of participants

Demographic characteristics and clinical motor outcome measures of our stroke participants indicate that they were chronic stroke survivors with mild-to-moderate motor impairment. Our stroke participants’ baseline demographics and clinical motor outcomes are summarized in Table 1. Mean age of the participants was 59.43 ± 12.57 years old, ranging from 32 to 80 years old. Participants’ mean chronicity was 3.01 ± 3.10 years after onset, ranging from 6 months to 14 years after onset. Further, participants had mild to moderate structural damage on the ipsilesional CST: the range of CST-lesion overlap volume was from 0% to 25% of the entire CST volume. The range of Upper Extremity Fugl-Meyer was from 25 to 58 out of 66, and the range of modified WMFT time score was from 0.82 to 4.71 second[log]).

Table 1.

Participants’ characteristics

Characteristics Mean (Standard deviation)
Age [years] 59.43 (12.57)
Sex [Male/Female] 27/20
Chronicity [years] 3.01 (3.10)
Hand dominance [Rt/Lt] 34/3
Affected hemisphere [Rt/Lt] 18/19
Lesion volume [mm3] 19840.89 (37868.69)
CST-lesion overlap volume [%] 4.59 (5.88)
Fugl Meyer Upper Extremity 43.49 (8.99)
WMFT time score – distal items [log(sec)] 2.20 (1.08)
Open in a new tab

3.2. Difference in CST FA between Ipsilesional and Contralesional tracts

All seven methods showed that the ipsilesional CST FA was lesser than the contralesional CST FA (Figure 2). The slice by slice analysis of 3-D individual CST tractography showed that in group level, the FA of ipsilesional CST was significantly decreased compared to the FA of contralesional CST only in the most damaged axial slices. On the other hand, the slice by slice analysis of two 3-D template CST methods showed that the FA of ipsilesional CST was significantly decreased in a majority of axial slices (Figure 3 A – C). The individual level slice by slice analysis of individual CST tractography revealed that only 15 of 37 (40%) participants had lower FA in the remote area to the CST-lesion junction, while the other ~60% showed lower FA of CST near the CST-lesion intersection only (Figure 3 C and D).

Figure 2.

Figure 2

Open in a new tab

Difference in FA between ipsi- and contra-lesional CSTs from different methods. (A) E1: 3D individual tractography. (B) E2: 3D atlas template. (C) E3: 3D subject template. (D) E4: 3D PLIC. (E) E5: 3D CP. (F) E6: 2D PLIC. (G) E7: 2D CP. (***P < .001 with Bonferroni correction)

Figure 3.

Figure 3

Open in a new tab

Comparison of FA between ipsi- and contra-lesional CST in axial slices. (A) Mean CST FA across all participants from 3-D individual CST tractography-based method (E1). (B) Mean CST FA across all participants from3-D template atlas CST-based method (E2). (C) Mean CST FA across all participants from 3-D template CST from participants’ contralesional CST tractography method. (D) A Participant who showed lower FA of ipsilesional CST than contralesional CST only near the CST-lesion intersection. (E) A participant who showed lower FA of ipsilesional than contralesional CST also in remote areas to CST-lesion intersection. Blue line – contralesional CST FA; Red line – ipsilesional CST FA; Black line - # of lesion voxels (right y-axis). Whiskers indiciate standard error. Green vertical dashed lines indicate the Cerebral Peduncle (CP) and Posterior Limb of the Internal Capsule (PLIC) sections of CST. The black dots on red line indicate the significant difference in FA between ipsi- and contra-lesional sides (p < .000276).

3.3. Difference in CST FA asymmetry among different methods

There was a statistically significant difference in CST FA asymmetry among the different methods (F stat = 4.88, p <0.001) (Table 3, Figure 4). Post-hoc analysis revealed that FA asymmetry from the 3-D CP method (E4) was significantly smaller than that from the CST atlas template (E2), 3-D PLIC (E5), and 2-D PLIC (E6) (Figure 4). Only the CST FA asymmetry for the 3-D individual CST tractography (E1) and 3-D subject Template CST methods (E3) fell within the normative CST FA asymmetry range. Maximum CST FA asymmetry from three of the seven methods (3-D CST atlas template [E2], 3-D PLIC [E4], and 2-D PLIC [E6]) was equal to or greater than 0.25. Additionally, the minimum CST FA asymmetry from the 3-D CP (E5) and 2-D CP (E7) methods was equal to or below −0.03 (Figure 4, Table 4).

Table 3.

ANOVA of FA Asymmetry.

Source SS df MS F-stat P value
Methods 0.11 6 0.018 4.88 <.001***
Error 0.94 252 0.004
Total 1.05 258
Open in a new tab
***

p < .001 with Bonferroni correction

Figure 4.

Figure 4

Open in a new tab

Comparison of FA asymmetry from different methods. Red horizontal dash line indicates the normative range of CST FA asymmetry in this cohort of chronic stroke survivors with mild to moderate motor impairment. (*p < .05, **p < .01, ***p < .001). E1 – 3-D individual CST tractography, E2 – 3-D atlas template CST; E3 – 3-D subject template CST; E4 – 3-D PLIC VOI; E5 – 3-D CP VOI; E6 – 2-D PLIC ROI; E7 – 2-D CP ROI.

Table 4.

Summary of Statistics of FA Asymmetry.

Methods Min Max Mean SD
E1: 3-D Tractography
CST VOI −0.02 0.14 0.06 0.04
E2: 3-D Atlas Template
CST VOI −0.03 0.29 0.09 0.08
E3: 3-D Subject
Template CST VOI −0.01 0.11 0.05 0.03
E4: 3-D PLIC VOI −0.06 0.28 0.08 0.08
E5: 3-D CP VOI −0.07 0.11 0.02 0.04
E6: 2-D PLIC ROI −0.05 0.25 0.07 0.08
E7: 2-D CP ROI −0.13 0.21 0.06 0.06
Open in a new tab

Bold indicates the values that are considered out of normative FA asymmetry boundary for our population.

3.4. Relationships between neuroimaging CST variables and WMFT-distal time score

There was a statistically significant partial correlation between 3-D individual CST tractography-based FA asymmetry (E1) and modified WMFT log-transformed mean time score for distal control items. No other significant correlations were found for any of the other DTI-derived CST methods (Figure 5, Table 5).

Figure 5.

Figure 5

Open in a new tab

Partial correlation between individual tractography-based CST FA asymmetry (E1) and mWMFT time score, controlled for age and chronicity.

Table 5.

Partial correlation coefficients between DTI-derived CST FA asymmetries and mWMFT time score.

E1 E2 E3 E4 E5 E6 E7
mWMFT 0.46* 0.05 0.18 0.14 −0.02 0.02 −0.19
Open in a new tab
*

p < .05 with Bonferroni correction.

E1 – 3-D individual CST tractography, E2 – 3-D atlas template CST; E3 – 3-D subject template CST; E4 – 3-D PLIC VOI; E5 – 3-D CP VOI; E6 – 2-D PLIC ROI; E7 – 2-D CP ROI.

4. Discussion

Our results provide robust evidence that among seven different DTI-derived estimates of CST, 3-D individual CST tractography-based FA asymmetry (E1) is the best measure for characterizing CST structure in this cohort of chronic stroke survivors with mild-to-moderate impairment. In the following discussion, we provide an explanation for this finding in light of the three criteria we used to evaluate the seven different methods:

First, all seven methods were able to detect the decreased fractional anisotropy in the ipsilesional CST compared to the contralesional CST. This result indicates that all seven methods met the first criteria for the appropriate CST structural estimates.

Second, based on previous reports (Lindenberg et al., 2012; Puig et al., 2013; Stinear et al., 2007), participants with motor impairment in the mild to moderate range, are expected to have CST FA asymmetry that is less than 0.25 and greater than −0.03. This lower limit is an estimation from previous studies that reported the FA of left and right CST in non-disabled adults (Lindenberg et al., 2012) and CST FA ratio between hemispheres in mildly impaired stroke survivors (Puig et al., 2013). CST FA asymmetry greater than 0.25 indicates that the individual has no recovery potential, which implies severe stroke damage on the descending motor pathways (Stinear et al., 2007). Only two methods, 3-D individual CST tractography (E1) and 3-D subject Template CST (E3) met our expected range boundary for this stroke cohort (Table 4). The other five methods either under- or over-estimated CST FA asymmetry. Therefore, with respect to FA asymmetry boundary range, the 3-D individual CST tractography-based method and the 3-D subject Template CST method appear to be the most accurate methods for quantification of CST structural characteristics in our population.

Third, only the 3-D individual CST tractography-based FA asymmetry was significantly correlated with our clinical motor behavior measure – the modified WMFT time score. We used partial correlation analysis to control for age and chronicity. It has been shown that age and chronicity has an impact on the relationship between brain biomarker and clinical motor outcomes (Cramer, 2008; Lindenberg et al., 2012; Stinear et al., 2007). We also conducted the pearson correlation analysis between CST FA asymmetry and the modified WMFT time score (data not shown). The result from correlation analysis was similar to the result from partial correlation analysis. This indicates that the age and chronicity did not affect the relationship between the CST FA asymmetry and motor outcome measure in this cohort of stroke survivors. It has been shown that the amount of CST structural damage is a major factor associated with motor deficits after stroke (Corbetta et al., 2015). Therefore, we expected the FA asymmetry index from an accurate quantification of CST damage would be significantly correlated with motor performance. We believe that the FA asymmetry index from the other six methods were not the most accurate representation of CST structural damage and therefore were not found to correlate significantly with WMFT-distal time score. Further, the fact that our results showed that a measure of CST structural impairment is associated with motor performance may be due to the vital role of CST in delivering precise motor commands for integrative functional upper extremity movements, especially those for distal arm and hand manipulation. Disruption of CST structure can disharmonize the motor commands leading to poor distal arm motor control (Dong et al., 2006; van Kuijk et al., 2009). Further, a previous study showed that WMFT functional assessment score (FAS) may have two different measurement domains: proximal control domain and integrative functional control domain (Woodbury et al., 2010). In addition, given that our participants were mild-to-moderately impaired in their motor performance (i.e. primarily distal upper limb motor deficits), there was a ceiling on the time scores for proximal control items. Most participants performed these proximal items on average within a few seconds, and with a similar time score to their less-affected side (Appendix 1). Therefore, a mean WMFT time score computed using all fifteen items is a poor estimate of the specific upper extremity motor deficit of this cohort. Further, clinical motor outcome measures related to distal arm and hand function may be more appropriate than general motor impairment measures, such as the Upper Extremity Fugl-Meyer, to determine the relationship between CST structure and motor behavior in chronic stroke survivors with mild-to-moderate motor impairment.

The most plausible reason why only the 3-D individual tractography-based CST FA asymmetry (E1) was significantly correlated with upper extremity motor performance is that other methods may not accurately estimate the ipsilesional CST microstructure. It has been shown that the degree of ipsilesional CST damage is correlated with upper extremity motor impairment and/or performance (Kim and Winstein, 2017; Lindenberg et al., 2010; Puig et al., 2013). If a DTI-derived CST estimate accurately represents CST microstructure, then this estimate would be expected to correlate strongly with motor performance. Therefore, we believe that the significant correlation (though not strong) between 3-D individual tractography-based CST FA asymmetry and motor capability, strongly suggests that this estimate represents the most accurate quantification of CST microstructure in this chronic cohort of stroke survivors. Another reason would be that methods based on a template identify the white matter ROIs by mapping from an atlas to each subject. This mapping is based on a registration of the atlas image to that of the subject. BrainSuite, and most other software including SPM, performs the registration using T1-weighted images which are good at aligning gray, white and cerebrospinal fluids (CSF) regions. However, since there is no significant contrast variation within white matter in T1-weighted images, these registration methods will generally not accurately align the CST or other fiber tracts. The tractography-based methods avoid this problem by using diffusion MRI based information to directly delineate CST. It is therefore not surprising that this approach proves more powerful in identifying stroke-related effects in CST.

In addition to the results that pertain to the three criteria, there are several reasons why we consider the individual tractography-based CST quantification to be the most appropriate estimate of CST structural characteristics.

Individual tractography-based CST quantification accounts for a distorted ipsilesional CST trajectory, whereas template-based CST quantification is unable to do so (Figure 6). The distorted ipsilesional CST trajectory is easily observed in chronic stroke survivors in large part due to significant subcortical white matter atrophy (Jang, 2011). Our tensor-based tractography method was able to accurately trace the CST streamlines in the ipsi-lesional hemisphere for all participants. However, this was not the case using the template method. Primarily due to white matter atrophy and enlarged ipsilesional lateral ventricle, the template-based method inaccurately included the lateral ventricle in 12 of 37 participants (i.e. over 1/3rd of the sample). Given that the FA value is relatively low in the ventricle (i.e., more isotropic diffusion of water molecules in the ventricle), this may cause an overestimation of the CST structural damage. For this reason, individual tractography-based CST quantification more accurately represents the ipsilesional CST structural characteristics compared with template-based methods.

Figure 6.

Figure 6

Open in a new tab

3-D Template CST VOI and 3-D individual CST VOI from tractography on the T1-weighted image. Green indicates CST mask from diffusion tensor-based tractography from one participant, and red indicates template CST mask registered to subject space. (A) Frontal slice showing the overall trajectories of CST tractography and CST atlas template. (B) 3-D CST masks on an axial slice.

Further, 3-D individual tractography-based methods represent the entire CST. The 2-D ROI-based methods, frequently used in the literature, do not represent the entire CST. The assumption behind these 2-D ROI-based methods is that Wallerian degeneration causes the decreased FA in the remote area to the CST-lesion junction, such as PLIC or CP. The FA of ipsilesional PLIC or CP is assumed to reflect the structural integrity of the entire CST. The slice-by-slice analysis of 3-D individual CST tractography demonstrated that the decreased FA of ipsilesional CST is observed only in the most damaged axial slices (Figure 3A). Further, we found that only 15 of 37 (40%) participants had decreased FA in the remote area to the CST-lesion junction, such as cerebral peduncle and posterior limb of the internal capsule (Figure 3D). The other ~60% showed decreased FA of CST near the lesional area only (Figure 3E). Thus, use of 3- D VOI or 2-D ROI of PLIC or CP may not capture the structural integrity of the entire CST for those who have no decrease in FA in these remote areas (e.g., Figure 3D).

Finally, individual CST tractography processing was automated with minimal user bias for computing the DTI-derived metrics. Although previous studies reported operator-dependent bias of CST tractography of stroke patients (Jang, 2011; Kwon et al., 2011), our CST tractography method using BrainSuite software was automated to a large extent with minimal to no operator manual correction necessary. In addition, the cortical regions for identifying CST were automatically labeled by the software, and subcortical structures, such as PLIC, CP and Pons, were manually drawn on a standard atlas and transformed to subject space to reduce inter-subject variability in ROIs. Thus, individual tractography-based CST quantification using BrainSuite is an objective method for determining CST structural characteristics with minimal operator-dependent bias.

The most common methods used in previous studies are including manually drawn 2-D posterior limb of the internal capsule (PLIC) and cerebral peduncle (CP) ROIs. The 2-D ROI-based estimate is the most simpe method, as it does not require any diffusion tensor tractography. In addition, template-based methods, including 3-D template CST, 3-D PLIC and CP VOIs, also did not require tractography. We believe these methods were popular as there was a difficulty in accurate diffusion tensor tractography. As now we have better hardware and software for tractography, tractography-based CST estimates are less challenging, and it can be done mostly automatically and less time consuming.

There has been only one previous work that compared different standard DTI-derived estimates of CST structure. Our results are inconsistent with the previous work by Park et al., (2013). Park and colleagues showed that different methods had similar correlation with motor ability, while our study showed only the individual tractography-bsed method has significant correlation with a motor outcome measure. There are several factors that may result in the inconcistent results between our work and the previous work: 1) difference in cohort of stroke survivors, 2) difference in DTI analysis software, 3) difference in CST template, and 4) difference in motor outcome measures.

First, our study recruited chornic stroke survivors with mild-to-moderate severity, while previous work was conducted with stroke surviors with broader range of severity and chronicity. The difference in cohort of stroke survivors may be a main reason of the inconsistent results between our work and previous work. Second, we used BrainSuite software to process MRI data, while Park and colleagues used Statistical Parmetric Mapping software (SPM, Wellcome Dept. of Cognitive Neurology, London, UK). BrainSuite software uses different co-registration algorithms and different tractography methods than does SPM. This difference in software may account for the inconsistent results between the two studies. Third, the previous work used the age-matched control participant’s CST tractography as their template, while our study used a template from standard atlas and another template generated from the contralesional hemisphere of our study participants. We assume that the contralesional CST structure is intact, and thus the template CST volume from contralesional CST tractography may be a better alternate for the template CST than that from age-matched non-disabled controls. Lastly, we utilized the WMFT-distal time score to represent the motor behavior of our participants. On the other hand, the previous work employed a composite value of different motor outcome measures derived from principal component analysis. Differences in clinical motor outcome measures between the two studies may have influenced the correlations between DTI-derived CST metrics and motor ability scores.

One limitation concerns the possibility of error in the spatial registration of stroke brain structural images. Distortions in the registered structural images could be introduced in case with enlarged ipsilesional lateral ventricle and image intensity changes in lesional and surrounding areas (Crinion et al., 2007; Feng et al., 2015). The consequence of the inaccurate spatial registration is that the template CST-derived estimates are inaccurate. Therefore, four of the seven estimate methods (3-D template CST from atlas, 3-D template CST from subjects, 3-D template PLIC and 3-D template PC) are directly impacted by this limitation. Another limitation is that our findings may not generalize to the wider stroke population. Given that we included chronic stroke survivors with mild-to-moderate severity, our findings are not applicable to stroke survivors at any stage of recovery with severe motor impairment, as tractography is limited in the presence of severe CST damage (Jang, 2011). Lastly, while our sample size is relatively large compared with previous DTI studies, ideally, we would need a larger sample to replicate and further validate our findings.

5. Conclusion

A systematic comparison of seven different DTI-derived estimates of CST structure found that 3-D individual tractography-based CST FA asymmetry is the most accurate estimate. The process of individual tractography-based CST FA computation using BrainSuite software is semiautomatic with minimal manual correction needed. As such, we recommend this method to highlight critical brain-behavior mechanisms for future stroke rehabilitation and recovery research.

Highlights.

  • Best estimate of CST microstructure for chronic stroke survivors with mild-to-moderate motor impairment is individual tractography-derived CST FA asymmetry.

  • Template-derived CST estimates do not account for shifted location of ipsilesional CST.

  • FA asymmetry at PLIC or CP do not represent the entire CST microstructure.

  • Individual tractography-derived CST FA asymmetry was the only estimate among the seven other DTI-derived CST estimates that showed a significant correlation with motor performance in mild-to-moderately impaired chronic stroke survivors.

Acknowledgments

This study was supported by the US National Institutes of Health, Eunice Kennedy Shriver National Institute of Child Health & Human Development (contract number: NIH/NINCHD HD065438).

Appendix 1. Box plots of WMFT Time scores

graphic file with name nihms963800f1.jpg

Open in a new tab

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

References

  1. Andersson JLR, Sotiropoulos SN. An integrated approach to correction for off-resonance effects and subject movement in diffusion MR imaging. Neuroimage. 2016;125:1063–78. doi: 10.1016/j.neuroimage.2015.10.019. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Bhushan C, Haldar JP, Choi S, Joshi AA, Shattuck DW, Leahy RM. Co-registration and distortion correction of diffusion and anatomical images based on inverse contrast normalization. Neuroimage. 2014;115:30–33. doi: 10.1016/j.neuroimage.2015.03.050. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Cho SH, Kim DG, Kim DS, Kim YH, Lee CH, Jang SH. Motor outcome according to the integrity of the corticospinal tract determined by diffusion tensor tractography in the early stage of corona radiata infarct. Neurosci Lett. 2007a;426:123–7. doi: 10.1016/j.neulet.2007.08.049. [DOI] [PubMed] [Google Scholar]
  4. Cho SH, Kim SH, Choi BY, Cho SH, Kang JH, Lee CH, Byun WM, Jang SH. Motor outcome according to diffusion tensor tractography findings in the early stage of intracerebral hemorrhage. Neurosci Lett. 2007b;421:142–6. doi: 10.1016/j.neulet.2007.04.052. [DOI] [PubMed] [Google Scholar]
  5. Corbetta M, Ramsey L, Callejas A, Baldassarre A, Hacker CD, Siegel JS, Astafiev SV, Rengachary J, Zinn K, Lang CE, Connor LT, Fucetola R, Strube M, Carter AR, Shulman GL. Common Behavioral Clusters and Subcortical Anatomy in Stroke. Neuron. 2015;85:927–941. doi: 10.1016/j.neuron.2015.02.027. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Cramer SC. Repairing the human brain after stroke: I. Mechanisms of spontaneous recovery. Ann Neurol. 2008;63:272–287. doi: 10.1002/ana.21393. [DOI] [PubMed] [Google Scholar]
  7. Crinion J, Ashburner J, Leff A, Brett M, Price C, Friston K. Spatial normalization of lesioned brains: Performance evaluation and impact on fMRI analyses. Neuroimage. 2007;37:866–875. doi: 10.1016/j.neuroimage.2007.04.065. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Dong Y, Dobkin BH, Cen SY, Wu AD, Winstein CJ. Motor cortex activation during treatment may predict therapeutic gains in paretic hand function after stroke. Stroke. 2006;37:1552–1555. doi: 10.1161/01.STR.0000221281.69373.4e. [DOI] [PubMed] [Google Scholar]
  9. Feng W, Wang J, Chhatbar PY, Doughty C, Landsittel D, Lioutas VA, Kautz SA, Schlaug G. Corticospinal tract lesion load: An imaging biomarker for stroke motor outcomes. Ann Neurol. 2015;78:860–870. doi: 10.1002/ana.24510. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Holmes CJ, Hoge R, Collins L, Woods R, Toga AW, Evans AC. Enhancement of MR images using registration for signal averaging. J Comput Assist Tomogr. 1998;22:324–33. doi: 10.1097/00004728-199803000-00032. [DOI] [PubMed] [Google Scholar]
  11. Jang SH. A review of diffusion tensor imaging studies on motor recovery mechanisms in stroke patients. NeuroRehabilitation. 2011;28:345–52. doi: 10.3233/NRE-2011-0662. [DOI] [PubMed] [Google Scholar]
  12. Joshi AA, Shattuck DW, Leahy RM. A Method for Automated Cortical Surface Registration and Labeling. Biomed image Regist proceedings WBIR (Workshop 2006-) 2012;7359:180–189. doi: 10.1007/978-3-642-31340-0_19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. 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. doi: 10.1177/1545968316662708. [DOI] [PubMed] [Google Scholar]
  14. Kim JH, Haldar J, Liang ZP, Song SK. Diffusion tensor imaging of mouse brain stem and cervical spinal cord. J Neurosci Methods. 2009;176:186–91. doi: 10.1016/j.jneumeth.2008.09.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Koyama T, Tsuji M, Miyake H, Ohmura T, Domen K. Motor outcome for patients with acute intracerebral hemorrhage predicted using diffusion tensor imaging: an application of ordinal logistic modeling. J Stroke Cerebrovasc Dis. 2012;21:704–11. doi: 10.1016/j.jstrokecerebrovasdis.2011.03.004. [DOI] [PubMed] [Google Scholar]
  16. Koyama T, Tsuji M, Nishimura H, Miyake H, Ohmura T, Domen K. Diffusion tensor imaging for intracerebral hemorrhage outcome prediction: comparison using data from the corona radiata/internal capsule and the cerebral peduncle. J Stroke Cerebrovasc Dis. 2013;22:72–9. doi: 10.1016/j.jstrokecerebrovasdis.2011.06.014. [DOI] [PubMed] [Google Scholar]
  17. Kuzu Y, Inoue T, Kanbara Y, Nishimoto H, Fujiwara S, Ogasawara K, Ogawa A. Prediction of motor function outcome after intracerebral hemorrhage using fractional anisotropy calculated from diffusion tensor imaging. Cerebrovasc Dis. 2012;33:566–73. doi: 10.1159/000338904. [DOI] [PubMed] [Google Scholar]
  18. Kwon HG, Hong JH, Jang SH. Anatomic location and somatotopic arrangement of the corticospinal tract at the cerebral peduncle in the human brain. AJNR Am J Neuroradiol. 2011;32:2116–9. doi: 10.3174/ajnr.A2660. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Lin JH, Hsu MJ, Sheu CF, Wu TS, Lin RT, Chen CH, Hsieh CL. Psychometric comparisons of 4 measures for assessing upper-extremity function in people with stroke. Phys Ther. 2009;89:840–850. doi: 10.2522/ptj.20080285. [DOI] [PubMed] [Google Scholar]
  20. Lindenberg R, Renga V, Zhu LL, Betzler F, Alsop D, Schlaug G. Structural integrity of corticospinal motor fibers predicts motor impairment in chronic stroke. Neurology. 2010;74:280–287. doi: 10.1212/WNL.0b013e3181ccc6d9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Lindenberg R, Zhu LL, Rüber T, Schlaug G. Predicting functional motor potential in chronic stroke patients using diffusion tensor imaging. Hum Brain Mapp. 2012;33:1040–1051. doi: 10.1002/hbm.21266. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Park CH, Kou N, Boudrias MH, Playford ED, Ward NS. Assessing a standardised approach to measuring corticospinal integrity after stroke with DTI. NeuroImage Clin. 2013;2:521–33. doi: 10.1016/j.nicl.2013.04.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Puig J, Blasco G, Daunis-I-Estadella J, Thomalla G, Castellanos M, Figueras J, Remollo S, van Eendenburg C, Sánchez-González J, Serena J, Pedraza S. Decreased corticospinal tract fractional anisotropy predicts long-term motor outcome after stroke. Stroke. 2013;44:2016–8. doi: 10.1161/STROKEAHA.111.000382. [DOI] [PubMed] [Google Scholar]
  24. Puig J, Pedraza S, Blasco G, Daunis-i-Estadella J, Prados F, Remollo S, Prats-Galino A, Soria G, Boada I, Castellanos M, Serena J. Acute Damage to the Posterior Limb of the Internal Capsule on Diffusion Tensor Tractography as an Early Imaging Predictor of Motor Outcome after Stroke. Am J Neuroradiol. 2011;32:857–863. doi: 10.3174/ajnr.A2400. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Sandor S, Leahy R. Surface-based labeling of cortical anatomy using a deformable atlas. IEEE Trans Med Imaging. 1997;16:41–54. doi: 10.1109/42.552054. [DOI] [PubMed] [Google Scholar]
  26. Shattuck DW, Sandor-Leahy SR, Schaper KA, Rottenberg DA, Leahy RM. Magnetic Resonance Image Tissue Classification Using a Partial Volume Model. Neuroimage. 2001;13:856–876. doi: 10.1006/nimg.2000.0730. [DOI] [PubMed] [Google Scholar]
  27. Stinear CM, Barber PA, Petoe M, Anwar S, Byblow WD. The PREP algorithm predicts potential for upper limb recovery after stroke. Brain. 2012;135:2527–2535. doi: 10.1093/brain/aws146. [DOI] [PubMed] [Google Scholar]
  28. Stinear CM, Barber PA, Smale PR, Coxon JP, Fleming MK, Byblow WD. Functional potential in chronic stroke patients depends on corticospinal tract integrity. Brain. 2007;130:170–80. doi: 10.1093/brain/awl333. [DOI] [PubMed] [Google Scholar]
  29. Thiebaut de Schotten M, Ffytche DH, Bizzi A, Dell’Acqua F, Allin M, Walshe M, Murray R, Williams SC, Murphy DGM, Catani M. Atlasing location, asymmetry and inter-subject variability of white matter tracts in the human brain with MR diffusion tractography. Neuroimage. 2011;54:49–59. doi: 10.1016/j.neuroimage.2010.07.055. [DOI] [PubMed] [Google Scholar]
  30. van Kuijk AA, Pasman JW, Hendricks HT, Zwarts MJ, Geurts ACH. Predicting hand motor recovery in severe stroke: the role of motor evoked potentials in relation to early clinical assessment. Neurorehabil Neural Repair. 2009;23:45–51. doi: 10.1177/1545968308317578. [DOI] [PubMed] [Google Scholar]
  31. Wolf SL, Catlin PA, Ellis M, Archer AL, Morgan B, Piacentino A. Assessing Wolf Motor Function Test as Outcome Measure for Research in Patients After Stroke. Stroke. 2001;32:1635–1639. doi: 10.1161/01.STR.32.7.1635. [DOI] [PubMed] [Google Scholar]
  32. Woodbury M, Velozo CA, Thompson PA, Light K, Uswatte G, Taub E, Winstein CJ, Morris D, Blanton S, Nichols-Larsen DS, Wolf SL. Measurement structure of the Wolf Motor Function Test: implications for motor control theory. Neurorehabil Neural Repair. 2010;24:791–801. doi: 10.1177/1545968310370749. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Yu C, Zhu C, Zhang Y, Chen H, Qin W, Wang M, Li K. A longitudinal diffusion tensor imaging study on Wallerian degeneration of corticospinal tract after motor pathway stroke. Neuroimage. 2009;47:451–8. doi: 10.1016/j.neuroimage.2009.04.066. [DOI] [PubMed] [Google Scholar]

RESOURCES