Can Neurological Biomarkers of Brain Impairment Be Used to Predict Poststroke Motor Recovery? A Systematic Review

Abstract

Background. There is growing interest to establish recovery biomarkers, especially neurological biomarkers, in order to develop new therapies and prediction models for the promotion of stroke rehabilitation and recovery. However, there is no consensus among the neurorehabilitation community about which biomarker(s) have the highest predictive value for motor recovery. Objective. To review the evidence and determine which neurological biomarker(s) meet the high evidence quality criteria for use in predicting motor recovery. Methods. We searched databases for prognostic neuroimaging/neurophysiological studies. Methodological quality of each study was assessed using a previously employed comprehensive 15-item rating system. Furthermore, we used the GRADE approach and ranked the overall evidence quality for each category of neurologic biomarker. Results. Seventy-one articles met our inclusion criteria; 5 categories of neurologic biomarkers were identified: diffusion tensor imaging (DTI), transcranial magnetic stimulation (TMS), functional magnetic resonance imaging (fMRI), conventional structural MRI (sMRI), and a combination of these biomarkers. Most studies were conducted with individuals after ischemic stroke in the acute and/or subacute stage (~70%). Less than one-third of the studies (21/71) were assessed with satisfactory methodological quality (80% or more of total quality score). Conventional structural MRI and the combination biomarker categories ranked “high” in overall evidence quality. Conclusions. There were 3 prevalent methodological limitations: (a) lack of cross-validation, (b) lack of minimal clinically important difference (MCID) for motor outcomes, and (c) small sample size. More high-quality studies are needed to establish which neurological biomarkers are the best predictors of motor recovery after stroke. Finally, the quarter-century old methodological quality tool used here should be updated by inclusion of more contemporary methods and statistical approaches.

Keywords

stroke motor recovery prognosis neurological biomarkers diffusion tensor imaging transcranial magnetic stimulation magnetic resonance imaging

Introduction

There is growing interest in establishing stroke recovery biomarkers. Researchers define stroke recovery biomarkers as surrogate indicators of disease state that can have predictive value for recovery or treatment response.¹ Specifically, previous studies have suggested that better understanding of neurological biomarkers, derived from brain imaging and neurophysiological assessments, is likely to move stroke rehabilitation research forward.^1,2

Recovery biomarkers acquired during the acute and subacute phases (acute—within 1 week after onset; subacute—between 1 week and 3 months after onset) may be vital to set attainable neurorehabilitation goals and to choose proper therapeutic approaches based on the recovery capacity. Furthermore, motor recovery prediction using neurological biomarkers in the chronic phase (more than 3 months after onset) can be useful to determine whether an individual will benefit from specific therapeutic interventions applied after the normal period of rehabilitation has ended. Hence, use of recovery biomarkers is likely to improve customization of physical interventions for individual stroke survivors regarding their capacity for recovery, and to facilitate development of new neurorehabilitation approaches.

There have been fundamental changes in recovery biomarkers from simple clinical behavioral biomarkers to brain imaging and neurophysiological biomarkers. In particular, a number of recent studies have shown that neurologic biomarkers (ie, neuroimaging and/or neurophysiological measures of brain) are more predictive of motor recovery than clinical behavioral biomarkers.^3-5

Although there is some evidence that neurological biomarkers are more valuable as predictors of motor recovery than clinical behavioral biomarkers, there are significant gaps between the published evidence and clinical usage. First, there is no consensus on which specific neurological biomarkers would be best for prediction models.^4,6,7 Viable neurological biomarker of motor recovery have evolved from lesion size and location, prevalent in the early 1990s⁸ to more contemporary complex brain network analysis variables.⁹ Despite this evolution, there is a paucity of high-level evidence for determining the most critical neurological biomarkers of motor recovery. A number of literature reviews and systematic reviews of studies published since the 1990s aimed to identify the most appropriate biomarkers of motor recovery or functional independence.^8,10-12 Among these reviews, only one by Schiemanck and colleagues⁸ assessed the evidence quality of neurologic biomarkers, while many focused on clinical measures (ie, clinical motor and/or functional measures).¹¹ Their review was limited to only 13 studies that employed structural magnetic resonance imaging (sMRI) measures of lesion volume as neurologic biomarkers. Besides lesion volume derived from structural MRI, there are other viable neurological biomarkers of brain impairment. Therefore, this systematic review includes a broad set of relevant biomarkers for consideration as critical predictors for inclusion in motor recovery prediction models.

Furthermore, there is some evidence to suggest that multivariate prediction models that use neurological biomarkers in addition to clinical outcome measures are more accurate than those that use clinical outcome measures alone.^2,13 However there is still no consensus about whether incorporating behavioral and neurological predictors in a multimodal prediction model is superior (ie, more accurate) to a univariate model that includes either behavioral or neurological predictors alone.

Taken together, this systematic review has 2 aims. The first is to conduct a critical and systematic comparison of selected studies to determine which neurological biomarker(s) is likely to have sufficient high-level evidence in order to render the most accurate prediction of motor recovery after stroke. The second aim is to identify whether adding clinical measures along with neurological biomarkers in the model improves the accuracy of the model compared to the models that use neurological biomarkers alone.

Methods

Inclusion and Exclusion Criteria

Given the goal to predict motor recovery after stroke, the inclusion and exclusion criteria were adapted from a recent systematic review of the same topic (Table 1).¹¹ Major differences are that Chen and Winstein¹¹ used the International Classification of Functioning and Disability as an organizing framework for dependent measures (ie, behavioral outcomes), and they included clinical prognostic studies without neuroimaging/neurophysiological predictors.

Table 1.

Inclusion and Exclusion Criteria.

Categories	Inclusion Criteria	Exclusion Criteria
Population characteristics	Postcerebral stroke with upper/lower limb deficits	Brainstem or cerebellar stroke
	Age older than 18 years	Nonadult population
	Sample size ≥5
Study design	Longitudinal study with different time points of predictor and outcome measures	Cross-sectional study
Measurement variables	Behavioral outcome (dependent) measures related to upper and/or lower limb recovery at 2 different time points	Outcome measures not specifically related to upper/lower limb recovery (eg, cognition, language etc)Outcome measures for specific upper/lower limb impairments (eg, pain, sensory deficit, or spasticity)
Measurement variables	Predictor (independent) measures using one or more neuroimaging/neurophysiological measures (eg, MRI, DTI, or TMS)	Predictor measures using behavioral measures only
Language	English language only	Non-English language

Abbreviations: DTI, diffusion tensor imaging; MRI, magnetic resonance imaging; TMS, transcranial magnetic stimulation.

Literature Search Strategy

Research articles published before December 2015 were searched for in PubMed, ISI Web of Knowledge, and Google Scholar. Search keywords included [“stroke” and “motor recovery” and “predict”] and one of the following keywords [“neuroimaging” or “neurophysiological measure” or “diffusion tensor imaging” or “magnetic resonance imaging” or “transcranial magnetic stimulation”].

Evidence Methodological Quality Evaluation

We evaluated 71 studies using the evidence methodological quality score (EQS). The methodological quality grading criteria were adapted from previous systematic reviews of a similar topic.^8,11,14,15 This evaluation system includes 3 categories: internal validity, statistical validity, and external validity. Each category contains several items, and each item is scored using a binary system: yes (1) or no (0). The maximum EQS is 15 (Table 2; see the appendix for detail).

Table 2.

Evidence Quality Score (EQS) Evaluation Categories.

	Internal validity
	● Appropriate operational definitions of outcome and predictor variables
A	○ Outcome measures
B	○ Predictor measures
	● Additional tests or citations for validity or reliability of measures
C	○ Outcome measures
D	○ Predictor measures
E	● Blinded evaluations
F	● Appropriate observation time points
G	● Control of dropout
	Statistical validity
H	● Control for statistical significance
I	● Appropriate sample size
J	● Control for multicollinearity
	External validity
K	● Stroke pathology identification
L	● Specification of inclusion and exclusion criteria
M	● Control for additional treatment effects on the outcome measures
N	● Cross-validation of the prediction model
O	● Discussion of minimal clinically important differences (MCID)

This evidence methodological quality scheme was developed based on the recommendations of the “Task Force on Stroke Outcome Research of Impairments, Disabilities and Handicap,”¹⁶ and the methodological guidelines for stroke outcome research are consistent with these criteria.^15,17

Overall Evidence Quality Evaluation

The evidence grading system from the GRADE Working Group¹⁸ was adapted to evaluate the overall evidence quality for each neurological biomarker category. We graded the overall evidence quality using a 4-level system: “High,” “Moderate,” “Low,” and “Very low.” Table 3 describes the criteria for each level.

Table 3.

Categories for Overall Evidence Quality for Each Biomarker Type.

Grade	Explanation
High	Further research is very unlikely to change our confidence in the estimate of the model
High	● More than five studies with higher than 12 (80% of maximum) methodological quality score, and average methodological quality score of the studies that used the neurological biomarker is greater than 10
Moderate	Further research is likely to have an important impact on our confidence in the estimate of the model and may change the estimate
	● Fewer than 5 studies with higher than 12 methodological quality score with consistent results, and average methodological quality score of the studies that used the neurological biomarker is greater than 10
	● More than 5 studies with higher than 12 methodological quality score, but average methodological quality score of the studies that used the neurological biomarker is less than 10
Low	Further research is very likely to have an important impact on our confidence in the estimate of the model and is likely to change the estimate
	● No study with a methodological quality score higher than 12, and average methodological quality score of the studies that used the neurological biomarker is greater than 10
	● Fewer than 5 studies with higher than 12 methodological quality score, and average methodological quality score of the studies that used the neurological biomarker is less than 10.
Very low	Any estimate of the model is very uncertain
	● No study with a methodological quality score higher than 12
	● Average methodological quality score of the studies is less than 10

Prediction Regression Model Evaluation

To evaluate these models, we estimated effect size and statistical power of each model. We assume that the statistical power of each prediction model reflects the accuracy and robustness of the model. If the article reported more than one model, each model’s statistical power and effect size were calculated separately. We extracted the R² value, significance level, and number of participants to calculate statistical power.^19,20 To compare the prediction models with or without clinical measures or demographic predictors, models were separated into 2 groups: (a) regression models using neurological biomarkers alone and (b) regression models using neurological biomarkers in conjunction with clinical (or demographic) measures.

Results

Literature Search

In all, 452 English language articles were found from the 3 databases. By screening the title and abstract, 81 articles were selected for review based on inclusion and exclusion criteria. Of the articles selected from the screening, 10 were excluded. Finally, 71 articles were included in the evidence quality evaluation (Figure 1). Details of included studies are summarized in Table 4.

Figure 1.

Evidence search strategy diagram.

Table 4.

Summary of Included Articles.

First Author	Year	Stroke Phase	Stroke Pathology	No. of Participants	Neurological Biomarker Type	Observation Period (Months)
Arac³⁹	1994	A	I, H	27	TMS	6
Borich³	2014	C	I	13	DTI	0.25
Burke²⁸	2015	SA and C	I, H	43	Combination	6
Chang⁵³	2015	SA	I	14	TMS	6
Cho⁵⁴	2007	SA	I	55	DTI	6
Cho⁵⁵	2007	SA	H	40	DTI	3
Coutts⁵⁶	2009	A	I	457	CT	2.5
Cramer⁴	2007	C	I	24	fMRI	1
Dawes⁵⁷	2008	C	I	18	Combination	6
Di Lazzaro⁵⁸	2010	A and SA	I	17	TMS	0.5
Dong⁵⁹	2006	C	N/S	8	fMRI	6
Escudero⁶⁰	1998	A	I	50	TMS	5
Feng⁶¹	2015	A	I	76	MRI	3
Feydy⁶²	2002	SA	I	14	fMRI	12
Feys⁵	2000	SA	I	64	TMS	6
Granziera⁹	2012	A	I	12	Combination	4.5
Grässel⁶³	2010	A and SA	I	11	DTI	3
Groisser³⁸	2014	A	I	10	DTI	30
Hand⁶⁴	2006	A	I	82	MRI	9
Hendricks⁶⁵	1997	A	I	47	TMS	26
Hendricks⁶⁶	2003	A and SA	I	43	TMS	6
Holtmannspötter⁶⁷	2005	N/S	I	62	Combination	3
Jang⁷	2010	SA	H	53	Combination	1
Jang⁶⁸	2005	A and SA	I, H	31	DTI	6
Jang⁶⁹	2004	A	I, H	17	fMRI	12
Jung³²	2013	SA	I	8	fMRI	1
Kang⁷⁰	2013	A	I	284	MRI	2.56
Karibe⁷¹	2000	A	H	28	MRI	1
Kim⁷²	2015	SA	I, H	49	DTI	3
Koh⁷³	2015	SA	I, H	140	MRI	1.25
Koski⁷⁴	2004	C	N/S	10	TMS	1
Koyama⁷⁵	2012	SA	H	15	DTI	1
Koyama⁷⁶	2013	SA	H	32	DTI	3
Kusano⁷⁷	2009	A	H	18	DTI	6
Kuzu⁷⁸	2012	A	H	23	DTI	1.5
Kwon⁷⁹	2012	SA	I	71	DTI	3
Lai⁸⁰	2007	A	I	28	Combination	0.43
Lai⁸¹	2015	C	I, H	72	TMS	8
Liggins⁸²	2013	A	I	110	MRI	12
Lindenberg⁴⁷	2012	C	I	15	DTI	3
Liu⁸³	2012	A	I	48	DTI	2
Loubinoux⁸⁴	2003	SA	I	9	fMRI	0.5
Ma⁸⁵	2011	A	I	50	MRI	4
Ma⁸⁶	2014	A	H	25	DTI	1.5
Maeshima⁸⁷	2013	SA	H	25	DTI	3
Marumoto⁸⁸	2013	C	I	14	Combination	6
Nascimbeni⁸⁹	2006	A	I	33	TMS	3
Nouri⁹⁰	2011	C	I	60	Combination	24
Nuutinen⁹¹	2006	A or SA	I	48	Combination	6
Piron⁹²	2005	SA	I	20	TMS	0.5
Puig³⁷	2011	A	I	60	Combination	4.5
Puig⁴⁰	2013	A or SA	I	70	DTI	0.5
Qiu⁹³	2008	A	H	8	DTI	3
Quinlan³⁵	2015		N/S	40	Combination	12
Rapisarda⁹⁴	1996	A	I	26	TMS	6.5
Rehme³¹	2011	A	I	12	fMRI	1
Rehme⁹⁵	2015	A or SA	I	21	Combination	5
Rickards⁹⁶	2012	C	I	14	MRI	3
Rosso⁹⁷	2010	A	I	79	Combination	6.25
Schiemanck⁹⁸	2006	A and SA	I	75	MRI	3
Stinear²	2012	SA	I	40	Combination	6
Stinear¹³	2007	C	N/S	21	Combination	1.5
Tao⁹⁹	2014	A	H	32	Combination	3
Timmerhuis¹⁰⁰	1996	A	I	50	Combination	12
van Kuijk¹⁰¹	2009	A	I	39	TMS	3
Várkuti³⁰	2013	SA and C	I, H	9	fMRI	1
Wang¹⁰²	2012	A or SA	H	27	DTI	6.5
Wöhrle¹⁰³	2003	A	N/S	13	Combination	1
Yoshioka¹⁰⁴	2007	A	H	17	DTI	3
Yu¹⁰⁵	2009	A	I	9	DTI	0.5
Zarahn¹⁰⁶	2011	A	I	30	fMRI	3

Abbreviations: A, acute; SA, subacute; C, chronic; CT, computed tomography;DTI, diffusion tensor imaging; fMRI, functional magnetic resonance imaging; MRI, conventional structural MRI; H, hemorrhage; I, infarct; N/S, not stated; TMS, transcranial magnetic stimulation.

Trend in Number of Publications by Year

The earliest study to meet our inclusion/exclusion criteria was published in 1994. Generally, there was an increasing trend in the number of publications over time (Figure 2). On average, every year between 1994 and 2015 there have been approximately 3 publications per year with a high of 9 in 2012.

Figure 2.

Trend in number of publications per year. Most predictive studies using neurological biomarkers published after 2000. This trend is consistent with the development of noninvasive brain neuroimaging/neurophysiological assessment tools.

Stroke Pathology and Phases

A total of 3215 post–cerebral stroke individuals participated in the 71 studies included. The classification of pathology and stroke phase is summarized in Table 5. Among these participants, 1,800 (56%) from 31 studies were in the acute phase of recovery, 649 (20.2%) from 16 studies were in the subacute phase, and 269 (9.6%) from 12 studies were in the chronic phase when the neuroimaging/neurophysiological assessments occurred.

Table 5.

Number of Subjects by Across Studies by Stroke Pathology Type and Timing.

Stroke Phase Timing	Stroke Pathology				Total
Stroke Phase Timing	Ischemic, n (%)	Hemorrhagic, n (%)	Both, n (%)	Not Reported, n (%)	Total
Acute	1592 (49.52)	151 (4.70)	44 (1.37)	13 (0.40)	1800 (55.99)
Subacute	295 (9.18)	165 (5.13)	189 (5.88)	0 (0.00)	649 (20.19)
Acute/subacute	285 (8.86)	27 (0.84)	31 (0.96)	0 (0.00)	343 (10.67)
Chronic	158 (4.92)	0 (0.00)	72 (2.24)	79 (2.46)	309 (9.61)
Subacute/chronic	0 (0.00)	0 (0.00)	52 (1.62)	0 (0.00)	52 (1.62)
Not reported	62 (1.93%)	0 (0.00)	0 (0.00)	0 (0.00)	62 (1.93)
Total	2392 (74.40)	343 (10.67)	388 (12.07)	52 (2.86)	3215

Most studies were conducted with ischemic stroke patients (2392 of 3215, 75%). Approximately 50% of patients were in an acute phase with ischemic brain damage (Table 5).

Types of Neurological Biomarkers

There were 5 main categories of neurological biomarkers: diffusion tensor imaging (DTI), transcranial magnetic stimulation (TMS), functional MRI (fMRI), structural MRI (sMRI), and a combination of these biomarkers. Eighteen studies utilized these biomarkers and other clinical measures together as predictors (Table 6)

Table 6.

Summary of 5 Types of Neurological Biomarkers.

	DTI	TMS	fMRI	MRI	CT	Combination	Total
No. of studies (with behavioral predictors)	21 (2)	13 (4)	9 (1)	9 (3)	1 (1)	18 (7)	71
No. of participants	612	462	131	859	457	694	3215
% of total participants	19.04	14.37	4.07	26.72	14.21	21.59

Abbreviations: CT, computed tomography; DTI, diffusion tensor imaging; fMRI, functional magnetic resonance imaging; TMS, transcranial magnetic stimulation.

In the acute phase, DTI, TMS, sMRI, and combination biomarker types were similarly employed for the prediction model. Functional MRI was the least frequently used type (3 of 31 studies with acute participants). In the subacute phase, DTI biomarker type was predominantly used (7 of 16 studies with subacute participants). In the chronic phase, combination type was the most frequent type (5 of 12 studies with chronic participants) (Figure 3).

Figure 3.

Frequency of each neurological biomarker type for each stroke phase.

As prevalence of CT biomarker was too low to discuss the methodological quality, we excluded the CT biomarker from the discussion (only 1 large-scale retrospective observational study used a CT biomarker).

We describe the details of predictor variables for each neurological biomarker type in the supplementary material.

Clinical Endpoints for Motor Recovery

There were approximately 35 different clinical measures to capture motor recovery after stroke (Figure 4). The most frequently (6 or more studies) utilized clinical endpoint measures were: National Institutes of Health Stroke Scale (NIHSS, including modified version), Rankin Scale (RS, including modified version), Barthel Index (BI), Fugl-Meyer Assessment (FMA, including upper and lower extremities, or separate versions), Motricity Index (MI), Medical Research Council (MRC) score, and Wolf Motor Function Test (WMFT) Time score.

Figure 4.

Summary of motor endpoint (dependent) measures. Most prediction studies used broadly defined clinical endpoints, such as the Rankin Scale or the National Institutes of Health Stroke Scale.

Evidence Methodological Quality

The results of the evidence methodological quality evaluation are summarized in Table 7. The mean EQS (mean ± SD = 9.79 ± 2.13) of all reviewed studies was relatively high (range 4-13; Figure 5). There were 21 out of 71 studies (30%) with an EQS score of 12 (80% of total score) or more, and these studies are highlighted in Table 7.

Table 7.

Summary of Evidence Methodological Quality Score Evaluation.^a

First Author	Year	Neurological Biomarker Type	Internal Validity							Statistical Validity			External Validity					Total Score
First Author	Year	Neurological Biomarker Type	A	B	C	D	E	F	G	H	I	J	K	L	M	N	O	Total Score
Arac³⁹	1994	TMS	1	1	1	1	0	1	1	0	0	0	1	0	0	0	0	7
Borich³	2014	DTI	1	1	1	0	0	0	1	1	0	1	1	1	1	0	0	9
Burke²⁸	2015	Combination	1	1	0	1	1	1	1	1	0	1	1	1	1	1	0	12
Chang⁵³	2015	TMS	1	1	1	1	0	1	1	0	0	0	1	1	0	0	0	8
Cho⁵⁴	2007	DTI	1	1	1	1	0	1	0	1	1	0	1	1	0	0	0	9
Cho⁵⁵	2007	DTI	1	1	1	1	0	1	0	1	0	0	1	1	0	0	0	8
Coutts⁵⁶	2009	CT	0	1	0	1	0	1	0	1	1	1	1	1	0	0	0	8
Cramer⁴	2007	fMRI	1	1	1	1	0	0	1	1	0	1	1	1	1	0	0	10
Dawes⁵⁷	2008	Combination	1	1	1	1	0	0	0	1	0	1	1	1	1	0	0	9
Di Lazzaro⁵⁸	2010	TMS	1	1	1	1	0	1	0	1	0	1	1	1	1	0	0	10
Dong⁵⁹	2006	fMRI	1	1	1	1	0	0	1	1	0	1	0	1	1	1	0	10
Escudero⁶⁰	1998	TMS	1	1	1	1	0	1	1	1	1	1	1	1	0	1	1	13
Feng⁶¹	2015	MRI	1	1	1	1	1	1	1	1	1	1	1	1	1	0	0	13
Feydy⁶²	2002	fMRI	1	1	1	1	0	1	0	0	0	0	1	1	0	0	0	7
Feys⁵	2000	TMS	1	1	1	1	0	1	1	1	1	1	1	1	0	0	0	11
Granziera⁹	2012	Combination	1	1	1	1	0	1	1	1	0	1	1	1	0	0	0	10
Grässel⁶³	2010	DTI	1	1	0	1	0	1	1	0	0	0	1	1	0	0	0	7
Groisser³⁸	2014	DTI	1	1	1	1	0	1	1	1	0	0	1	1	0	0	0	9
Hand⁶⁴	2006	MRI	1	1	1	1	1	1	1	1	1	1	1	1	0	0	0	12
Hendricks⁶⁵	1997	TMS	1	1	1	1	0	1	1	0	0	0	1	1	0	0	0	8
Hendricks⁶⁶	2003	TMS	1	1	1	1	1	1	1	1	0	0	1	1	0	0	0	10
Holtmannspötter⁶⁷	2005	Combination	1	1	0	1	0	0	1	1	1	1	1	0	0	0	0	8
Jang⁷	2010	Combination	1	1	1	1	1	1	1	1	1	0	1	1	0	0	0	11
Jang⁶⁸	2005	DTI	1	1	0	1	1	1	1	1	0	1	1	1	0	0	0	10
Jang⁶⁹	2004	fMRI	1	1	0	1	0	1	1	0	0	0	1	0	0	0	0	6
Jung³²	2013	fMRI	1	1	0	1	0	1	1	0	0	0	1	1	0	0	0	7
Kang⁷⁰	2013	MRI	1	1	1	1	1	1	1	1	1	1	1	1	0	0	0	12
Karibe⁷¹	2000	MRI	1	1	0	0	0	0	1	0	0	0	1	1	0	0	0	5
Kim⁷²	2015	DTI	1	1	0	0	1	1	1	0	0	0	1	1	0	0	0	7
Koh⁷³	2015	MRI	1	1	0	0	0	0	1	1	1	1	1	1	0	0	0	8
Koski⁷⁴	2004	TMS	1	1	1	1	0	0	1	1	0	0	0	1	1	0	1	9
Koyama⁷⁵	2012	DTI	1	1	1	1	1	1	1	1	0	1	1	1	1	0	1	13
Koyama⁷⁶	2013	DTI	1	1	1	1	1	1	1	1	0	1	1	1	1	0	1	13
Kusano⁷⁷	2009	DTI	1	1	1	1	0	1	1	1	0	0	1	1	0	0	0	9
Kuzu⁷⁸	2012	DTI	1	1	1	1	0	1	1	1	0	1	1	1	0	0	0	10
Kwon⁷⁹	2012	DTI	1	1	1	1	1	1	1	1	1	1	1	1	0	0	0	12
Lai⁸⁰	2007	Combination	1	1	0	1	0	1	1	0	0	0	1	0	0	0	0	6
Lai⁸¹	2015	TMS	1	1	1	1	1	1	1	1	1	0	1	1	1	0	0	12
Liggins⁸²	2013	MRI	1	1	1	1	1	1	1	1	1	1	1	1	1	0	0	13
Lindenberg⁴⁷	2012	DTI	1	1	1	1	1	0	1	1	0	1	1	1	1	0	0	11
Liu⁸³	2012	DTI	1	1	0	1	0	1	1	1	0	1	1	1	0	0	0	9
Loubinoux⁸⁴	2003	fMRI	1	1	1	1	0	1	1	1	0	1	1	1	1	1	0	12
Ma⁸⁵	2011	MRI	1	1	1	1	0	1	1	1	1	1	1	1	0	0	0	11
Ma⁸⁶	2014	DTI	1	1	1	0	1	1	1	1	0	1	1	1	1	0	0	11
Maeshima⁸⁷	2013	DTI	1	1	1	1	1	1	1	1	0	1	1	1	1	0	0	12
Marumoto⁸⁸	2013	Combination	1	1	1	1	1	0	1	1	0	1	1	1	1	0	0	11
Nascimbeni⁸⁹	2006	TMS	1	1	0	1	0	1	1	1	0	1	1	1	1	0	0	10
Nouri⁹⁰	2011	Combination	1	1	0	1	1	0	1	1	1	1	1	1	1	0	1	12
Nuutinen⁹¹	2006	Combination	1	1	1	0	1	1	1	1	0	1	1	1	0	0	0	10
Piron⁹²	2005	TMS	1	1	1	1	1	1	1	1	0	0	1	1	0	0	0	10
Puig³⁷	2011	Combination	1	1	1	1	1	1	1	1	1	1	1	1	0	0	0	12
Puig⁴⁰	2013	DTI	1	1	1	1	1	1	1	1	1	1	1	1	0	0	0	12
Qiu⁹³	2008	DTI	1	1	0	0	0	1	1	1	0	0	1	0	0	0	0	6
Quinlan³⁵	2015	Combination	1	1	1	1	1	0	1	1	0	1	0	1	1	1	1	12
Rapisarda⁹⁴	1996	TMS	1	1	1	1	0	0	1	1	0	0	1	1	0	0	0	8
Rehme³¹	2011	fMRI	1	1	1	1	0	1	1	1	0	1	1	1	0	0	0	10
Rehme⁹⁵	2015	Combination	1	1	1	1	0	1	1	1	0	0	1	1	1	1	1	12
Rickards⁹⁶	2012	MRI	1	1	1	0	0	0	1	1	0	0	1	1	1	0	0	8
Rosso⁹⁷	2010	Combination	1	1	0	0	0	1	1	1	1	1	1	1	1	0	0	10
Schiemanck⁹⁸	2006	MRI	1	1	1	1	1	1	1	1	1	1	1	1	0	0	0	12
Stinear²	2012	Combination	1	1	1	1	1	1	1	1	0	0	1	1	1	0	1	12
Stinear¹³	2007	Combination	1	1	1	1	1	0	0	1	0	1	0	1	1	0	0	9
Tao⁹⁹	2014	Combination	1	1	1	1	1	0	1	1	0	0	1	1	0	0	0	9
Timmerhuis¹⁰⁰	1996	Combination	1	1	1	1	0	1	1	1	1	1	1	1	1	0	0	12
van Kuijk¹⁰¹	2009	TMS	1	1	1	1	1	1	1	1	0	0	1	1	1	0	1	12
Várkuti³⁰	2013	fMRI	1	1	0	1	0	1	1	1	0	1	0	1	1	1	0	10
Wang¹⁰²	2012	DTI	1	1	1	1	0	1	1	1	0	0	1	1	0	0	0	9
Wöhrle¹⁰³	2003	Combination	1	1	0	1	0	0	1	0	0	0	0	0	0	0	0	4
Yoshioka¹⁰⁴	2007	DTI	1	1	0	1	0	1	1	0	0	0	1	1	0	0	0	7
Yu¹⁰⁵	2009	DTI	1	1	1	1	0	1	1	1	0	0	1	1	0	1	0	10
Zarahn¹⁰⁶	2011	fMRI	1	1	1	0	0	1	1	1	0	1	1	1	0	0	0	9
		Total	70	71	52	61	28	54	64	59	20	41	65	65	28	8	9

Abbreviations: CT, computed tomography; DTI, diffusion tensor imaging; fMRI, functional magnetic resonance imaging; TMS, transcranial magnetic stimulation.

Shaded rows indicate the studies with high evidence quality (EQS ≥ 12).

Figure 5.

Distribution of evidence methodological quality scores.

Internal Validity

Most studies provided proper operational definitions of predictor and outcome measure variables. Fifty-two studies (73%) described validity and/or reliability of clinical endpoint measures, and 61 studies (86%) explained validity and reliability of their neurological measures. In 28 studies (39%), raters of predictor and outcome variables were blinded to the study purpose and other measures. In 54 studies (76%), the predictor variables were measured within 1 month after the index stroke, and the outcome measures were assessed at least 8 weeks after the measure of predictor variables. Among the 59 studies with poststroke individuals in the acute and subacute phases, the mean observation period was about 5 months after stroke (Table 8). Sixty-four studies (90%) described the number of and reasons for dropouts, or had no dropouts.

Table 8.

Summary of Observation Periods.

Observation Period	No. of Studies (% of Total)
<3 months	23 (32.4)
>3 months and <6 months	22 (31.0)
>6 months and <12 months	18 (25.4)
>12 months	8 (11.3)
Total	71

Statistical Validity

Most studies (59 of 71) applied appropriate statistical analyses. For the appropriate sample size, however, only 20 studies (~28%) met the appropriate sample size criteria. Furthermore, 41 studies (~58%) considered multicollinearity to control the effects of confounding variables on their correlation or regression analyses.

External Validity

Most studies (67 of 71) identified stroke pathology. Sixty-five studies (~92%) described inclusion and exclusion criteria. Six studies did not specify inclusion and exclusion criteria, or their descriptions of the criteria were insufficient to replicate the recruitment criteria. Only 28 studies (39%) discussed the effects of additional treatment on outcomes. Most of these studies described the treatments, such as physical therapy and/or occupational therapy, which were provided during study participation. Only 8 studies (~11%) performed the cross-validation of their prediction model using an independent group of participants with stroke. Nine studies (~13%) discussed clinically meaningful differences of predictors or outcome measures.

Overall Evidence Quality for Each Neurologic Biomarker Category

Table 9 summarizes the results of the overall quality grading. Conventional structural MRI and Combination biomarker types met the “High” grade criteria. DTI biomarker type met the “Moderate” grade criteria. TMS and fMRI met only the “Low” grade. The overall quality grade for the CT predictor was “Very low,” because there was only 1 study, which had a limited EQS score of eight.

Table 9.

Summary of Overall Evidence Quality of Each Biomarker Type.

Biomarker Type	No. of Studyies with EQS ≥ 12 (% of Total)	Average ± SD of EQS	Predictor Quality Category
Combination	7 (38.89)	10.06 ± 2.29	High
sMRI	5 (55.57)	10.44 ± 2.80	High
DTI	5 (23.81)	9.67 ± 2.03	Moderate
TMS	3 (23.08)	9.85 ± 1.82	Low
fMRI	1 (11.11)	9.00 ± 1.94	Low
CT	0 (0.00)	8.00 ± 0.00	Very low

Abbreviations: CT, computed tomography; DTI, diffusion tensor imaging; EQS, evidence methodological quality score; fMRI, functional magnetic resonance imaging; sMRI, structural magnetic resonance imaging; TMS, transcranial magnetic stimulation.

Prediction Regression Models

Among 71 reviewed studies, 32 (~45%) conducted linear or nonlinear regression analyses to develop prediction regression models that included neurological biomarkers. Thirty studies (~42%) reported statistically significant neurological biomarkers as predictors in their regression models. Among these 30 studies, 22 (~31%) informed statistical details of their regression models.

From the 22 studies that reported details, 39 regression models were identified that used neurological biomarkers. Among these models, 20 (~50% of reported models) consisted of neurological biomarkers and clinical measures as statistically significant predictors (Table 10).

Table 10.

Summary of Regression Models.^a

Reference	Predictors	Outcome Measures	Predictors From Clinical Measures	R ²	Effect Size (f²)	Statistical Power
Borich³	Chronicity, age, CST FA (ipsilesional and contralesional)	Changes in motor performance	Yes	0.75	2.94	0.97
Borich³	Chronicity, age, rFA of CST	Changes in motor performance	Yes	0.57	1.34	0.81
Burke²⁸	Leg FM Score, fMRI activation volume within ipsilesional foot primary sensorimotor cortex	Change in gait velocity from baseline to week 12	Yes	0.63	1.70	0.99
Cramer⁴	Arm motor FM score, degree of activation in ipsilesional primary motor cortex	Change in arm motor FM score baseline to week 10	Yes	0.40	0.67	0.93
Dawes⁵⁷	maximal cross-sectional overlap (degree of overlap of lesion with CST in mm), 10 m walk time, age	Walking speed	Yes	0.98	40.67	1.00
Dawes⁵⁷	maximal cross-sectional overlap (degree of overlap of lesion with CST in mm), 10m walk time, age	Normalized swing time asymmetry	Yes	0.99	70.43	1.00
Dong⁵⁹	LI of M1	Posttherapeutic change in the mean WMFT time	No	0.46	0.85	0.59
Escudero⁶⁰	BI, strength, CNS, CCT	BI	Yes	0.29	0.41	0.95
Escudero⁶⁰	BI, strength, CNS, CCT	Strength index	Yes	0.54	1.17	0.99
Feng⁶¹	Weighted CST-lesion load	FM-UE at 3 months after stroke	No	0.64	1.78	1.00
	Weighted CST-lesion load	NIHSS arm motor score at 3 months after stroke	No	0.58	1.38	1.00
	Weighted CST-lesion load	NIHSS total score at 3 months after stroke	No	0.45	0.82	1.00
Feys⁵	Motor performance, MEP, overall disability, and muscle tone within 1 month poststroke	Brunnstrom-Fugl-Meyer Score at 6 months	Yes	0.85	5.67	1.00
	Motor performance, SSEP at 2 months poststroke	Brunnstrom-Fugl-Meyer Score at 6 months	Yes	0.62	1.53	1.00
	Motor performance, MEP, overall disability, and muscle tone within 1 month poststroke	Brunnstrom-Fugl-Meyer Score at 12 months	Yes	0.82	4.14	1.00
	Motor performance, SSEP at 2 months poststroke	Brunnstrom-Fugl-Meyer Score at 12 months	Yes	0.58	1.38	1.00
Granziera⁹	NIHSS score, GFA M1-SMA, GFA PMd-PMv, GFA SMA-SMA, Age	NIHSS at 6 months poststroke	Yes	0.98	49.00	1.00
Holtmannspötter⁶⁷	MD	Change in Rankin Score about 26 months poststroke	No	0.36	0.56	0.99
Jang⁶⁸	rFA (CR group)	Change in MRC score 3 months poststroke	No	0.84	5.29	1.00
Jang⁶⁸	rFA (IC group)	Change in MRC score 3 months poststroke	No	0.57	1.31	0.98
Koh⁷³	Baseline UE motor score, stroke pathology type, baseline NIHSS score, lesion location	UE motor score at discharge	Yes	0.59	1.44	1
Koh⁷³	Stroke pathology type, baseline NIHSS score, lesion location	Change in UE motor score	Yes	0.47	0.87	1
Koyama⁷⁵	rFA	MRC grade (UE)	No	0.27	0.37	0.60
Koyama⁷⁵	rFA	MRC grade (LE)	No	0.25	0.33	0.55
Koyama⁷⁶	rFA of CP	mRS	No	0.23	0.31	0.87
	rFA of CP	MRC grade (UE)	No	0.27	0.37	0.92
	rFA of CP	MRC grade (LE)	No	0.19	0.24	0.77
Lindenberg⁴⁷	1. CST FA and M1-M1 FA	WMFT change (%) posttreatment (12 d from baseline)	No	0.61	1.56	0.98
	2. CST AD and M1-M1 AD	WMFT change (%)	No	0.64	1.78	0.99
	3. CST RD and M1-M1 AD	WMFT change (%)	No	0.69	2.23	0.99
Marumoto⁸⁸	PLIC rFA	FM-UE after treatment	No	0.65	1.86	0.99
Nascimbeni⁸⁹	NIHSS, MEP amplitude (percentage of compound motor action potential amplitude)	MRC score	Yes	0.74	3.00	0.99
Nascimbeni⁸⁹	NIHSS, MEP amplitude (percentage of compound motor action potential amplitude), MI	MRC score	Yes	0.75	2.82	0.99
Puig⁴⁰	1. PLIC damage	m-NIHSS 90 days poststroke	No	0.76	6.14	1.00
Puig⁴⁰	2. m-NIHSS and PLIC damage	m-NIHSS 90 days poststroke	Yes	0.86	9.00	1.00
Quinlan³⁵	iM1 activation, iM1-cM1 connectivity, Percentage injury to CST	Treatment-induced behavioral gains (from changes in FMA and ARAT)	No	0.44	0.79	0.97
Stinear¹³	FAAI, MEP presence, clinical score	Change in FMA score	Yes	0.91	10.11	1.00
Timmerhuis¹⁰⁰	MEP, BI, age	BI 6 weeks poststroke	Yes	0.73	2.70	0.99
Zarahn¹⁰⁶	Weighted CST-lesion load, raw CST-lesion load, lesion size	Change in FMA score at 3 months poststroke	No	0.73	2.70	1.00

Abbreviations: AD, axial diffusivity; ADC, apparent diffusion coefficient; ARAT, Action Research Arm Test; BI, Barthel Index; CCT, central conduction time; cM1, contralesional primary motor cortex; CNS, Canadian Neurological Scale; CST, corticospinal tract; DTI, diffusion tensor imaging; DWI, diffusion weighted imaging; FA, fractional anisotropy; FAAI, FA asymmetry index; FM, Fugl-Meyer assessment; GFA, generalized fractional anisotropy; iADC, ipsilesional hemisphere ADC; iFA, ipsilesional hemisphere FA; iM1, ipsilesional primary motor cortex; LI, Laterality Index; M1, primary motor cortex; MD, mean diffusivity; MEP, motor-evoked potential; m-NIHSS, modified NIHSS; MRC, Medical Research Council; mRS, Modified Rankin Score; NIHSS, National Institute of Health Stroke Scale; PLIC, posterior limb of internal capsule; PMd, premotor cortex dorsal part; PMv, premotor cortex ventral part; rADC, ratio of ADC between 2 hemispheres; RD, radial diffusivity; rFA, ratio of FA between ipsi- and contralesional hemispheres; SMA, supplementary motor area; UE, upper extremity; WMFT, Wolf Motor Function Test.

Effect size (Cohen’s f²) = R²/(1 − R²). Statistical power was calculated using R², number of predictors, probability level, and sample size. Shaded row indicates the models with low to moderate effect size (ie, Cohen’s f² < 0.8).

The mean statistical power and the mean effect size of the 39 prediction models was high (ie, 0.944 and 6.197, respectively). The statistical power and the effect size of multivariate models (ie, models using neurological biomarkers and clinical measures as predictors) were significantly greater than the power of the models using neurological biomarkers alone (Figures 6 and 7; Table 11). There were 4 studies that overestimated effect size (effect size is greater than 10), but removing these 4 studies did not influence the statistical results for comparing the effect sizes.

Figure 6.

Comparison of statistical power and effect size between models using neurological biomarkers alone and models using neurological biomarkers and clinical measures. (A) Statistical power; (B) effect size. Model group 1 indicates the models using neurological biomarkers alone, and the model group 2 indicates the models using neurological biomarkers and clinical measures as predictors. The extremely high effect size for model group 2 (see Table 10) has been removed from Figure 6B for plotting purposes.

Figure 7.

Forest plot for comparing effect sizes between different model groups. Each line represents 95% of confidence interval of effect size of each model. The size of black box represents sample size of each model. The diamonds indicate the mean of effect sizes for each model group. The effect size of models using neurological and clinical predictors were greater than that of models using neurological predictor alone.

Table 11.

Statistical Power and Effect Size Comparison Between Regression Models Using Neurological Biomarkers Alone and Models Using Neurological + Clinical (Demographic) Measures as Predictors.

	Mean	SD	Degrees of Freedom	t Statistic	P	Effect Size of t Test (Cohen’s d)
Statistical power comparison
Models with neurological biomarkers alone	0.9047	0.1557	37	2.0960	.043	0.24
Models with neurological biomarker and clinical measures	0.9807	0.0439
Effect size (Cohen’s f²) comparison
Models with neurological biomarkers alone	1.6143	1.6188	37	2.0115	.052	2.76
Models with neurological biomarker and clinical measures	10.5505	19.2872

Furthermore, 6 studies reported models with clinical motor behavioral predictors alone, in addition to the models incorporating behavioral and neurological predictors. Among these 6 studies, 5 reported that the prediction model with clinical behavioral predictors and neurological biomarkers explained more variance in the outcome variable than the model with clinical behavioral predictors alone (Table 12).

Table 12.

Prediction Models Using Only Clinical Behavioral Predictors.

Author	Predictors	Outcome Measures	R ²
Borich³	Poststroke duration, age	Changes in motor performance	0.008
Cramer⁴	Arm motor Fugl-Meyer score at baseline	Changes in arm motor Fugl-Meyer score	0.21
Feys⁵	Motor performance, muscle tone at baseline	Brunnstrom-Fugl-Meyer score at 2 months	0.74
Koh⁷³	Baseline UE motor score	Changes in UE motor score	0.04
	Baseline NIHSS score	Changes in UE motor score	0.05
	Baseline finger extension	Changes in UE motor score	0.06
Puig⁴⁰	Modified NIHSS within 3 days after stroke	Modified NIHSS at 90 days	0.90
Timmerhuis¹⁰⁰	Barthel index	Barthel index at 6 weeks poststroke	0.34
	Barthel index, age	Barthel index at 6 weeks poststroke	0.71

Abbreviations: NIHSS, National Institutes of Health Stroke Scale; UE, upper extremity.

Model in shaded row indicates model superiority compared with models using neurologic biomarkers.

Discussion

To our knowledge, this is the first evidence-based review to have critically and systematically evaluated the extant literature related to neurological biomarkers to determine the best predictor variables for motor recovery after stroke.

Evidence Methodological Quality

Numerous methodologically robust clinical studies provide evidence that structural biomarkers or a combination of different neurological biomarkers including DTI are useful to predict motor recovery after stroke. Several methodological weaknesses were found in studies using fMRI or TMS biomarkers, which included small sample size, a lack of blinded evaluation of outcome measures, no control for multicollinearity, or no control for additional treatment effects.

Furthermore, most studies (~90%) provided no cross-validation of the predictive models and no discussion of the minimal clinically important differences (MCID). A cross-validation of prediction models on an independent group of participants should be conducted to verify the validity and accuracy of prediction models.¹⁵ Therefore, the lack of cross-validation of the model would be the biggest limitation of the current literature. In recent studies, a number of statistical methods for model validation have been suggested, such as leave-one-out cross-validation or k-fold cross-validation.^21,22 Use of these statistical cross-validation methods is likely to improve the accuracy estimation of the prediction model.²¹ MCID is considered to be an important factor, particularly for interpretation of the relevance of observed changes in clinical endpoints.²³ Consideration of MCID for neurological biomarkers and clinical motor endpoints will likely improve the clinical usability of the predictive models for motor recovery.

Overall Evidence Quality

Only 2 types of biomarkers (ie, conventional sMRI biomarker type and combination type) were graded as high in overall evidence quality. Therefore, it is likely that we have sufficient evidence to utilize these 2 types of neurological biomarkers for development of prediction model.

Although DTI-derived biomarkers are the most frequently used in reviewed studies, the evidence methodological quality of those studies was insufficient. DTI is a promising noninvasive neuroimaging tool that captures orientation and microstructural characteristics of white matter in the human brain.²⁴ The popularity of DTI among stroke rehabilitation researchers is likely the ease with which one can quantify the structural characteristics of specific pathways affected by the stroke.²⁵ Therefore, future DTI studies that employ higher methodological quality are likely to have an important impact on our confidence in the prediction model(s) derived from them.

TMS measures were the second most prevalent biomarker (ie, 13 of 71 studies). However, only 3 of the 12 TMS studies had high methodological quality. The low prevalence of high-quality prognostic studies in this category resulted in a “low” overall evidence quality grade.

There was also a low prevalence of high-quality prognostic studies using fMRI. Specifically, only 1 fMRI study had an EQS ≥12. Thus, more methodologically robust prognostic studies using TMS or fMRI will be needed to raise our confidence in the estimate of the prediction model using these biomarkers.

Frequently Used Predictor Variables for Each Biomarker Type

In the following section, we will discuss the most frequently used predictor variables for each biomarker type. The details of predictor variables are described in the supplementary material.

DTI Biomarker Type

Ratio and asymmetry index of fractional anisotropy (FA) between ipsi- and contralesional corticospinal tracts (CSTs) were the most popular predictor variables in DTI studies among many DTI-derived variables in Table 13. FA of ipsilesional CST is associated with microstructural characteristics of white matter fibers.^3,26 A lower FA value of the ipsilesional CST may indicate greater damage of the CST that can lead to more Wallerian degeneration of CST axons.²⁷ However, the FA value can be influenced by a number of other factors, such as white matter architecture. Therefore, we need to use the DTI-derived FA values as neurologic biomarkers of brain impairment with caution.

Table 13.

Summary of DTI-Derived Predictor Variables.

Reference	DTI-Derived Predictor Variables	DTI Method	ROIs	Data Type
Borich³	iFA	2-D ROI on an axial slice	PLIC	Continuous
Cho⁵⁴	CST continuity	Tractography	CST	Categorical
Cho⁵⁵	CST continuity	Tractography	CST	Categorical
Granziera⁹	Generalized FA	Tractography	Motor network	Continuous
Grässel⁶³	ADC, rFA, rAD, rRD	3-D VOI	PT	Continuous
Groisser³⁸	dFA, dMD, dAD, dRD	Template3-D VOI	CST	Continuous
Jang⁶⁸	rFA	2-D ROI on an axial slice	CR and PLIC	Continuous
Jang⁷	CST continuity	Tractography	CST	Categorical
Kim⁷²	iFA, rFA, no. of fibers	Tractography	CST	Continuous
Koyama⁷⁵	rFA	2-D ROI on an axial slice	CP	Continuous
Koyama⁷⁶	rFA	Template3-D VOI	CST (CR/IC portion and CR portion separately)	Continuous
Kusano⁷⁷	rFA	2-D ROI on an axial slice	CP	Continuous
Kuzu⁷⁸	iFA	2-D ROI on an axial slice	CP	Continuous
Kwon⁷⁹	CST continuity	Tractography	CST	Categorical
Lai⁸⁰	CST Topography	Tractography	CST	Categorical
Lindenberg⁴⁷	iFA, iAD, iRD	Tractography	CST and aMF	Continuous
Liu⁸³	ADC and FA (ipsi, contra, and ratio)	2-D ROI on an axial slice	CP	Continuous
Ma⁸⁶	Generalized FA	2-D ROI on an axial slice	CP	Continuous
Maeshima⁸⁷	iFA	2-D ROI on an axial slice	CP	Continuous
Marumoto⁸⁸	rFA	Template3-D VOI	PLIC	Continuous
Puig³⁷	rFA	Tractography	CST at the pons	Continuous
Puig⁴⁰	Damage to specific CST regions	Tractography	MC, PMC, CS, CR, PLIC	Categorical
Qiu⁹³	Pyramidal fiber tract drawn lines	Tractography	CST (PT)	Continuous
Quinlan³⁵	iFA	2-D ROI on an axial slice	CP	Continuous
Rosso⁹⁷	ADC	Template3-D VOI	CST	Continuous
Stinear²	FA asymmetry index	Template3-D VOI	PLIC	Continuous and categorical
Stinear¹³	FA and MD asymmetry index	2-D ROI on an axial slice	PLIC	Continuous and categorical
Tao⁹⁹	rFA	2-D ROI on an axial slice	CP	Continuous
Wang¹⁰²	rFA	2-D ROI on an axial slice	CP	Continuous
Yoshioka¹⁰⁴	rFA and rADC	Tractography	CST	Continuous
Yu¹⁰⁵	rFA, rMD, rAD, rRD	Tractography	CST and CC	Continuous

Abbreviations: 2-D ROI, 2-dimensional region of interest; 3-D VOI, 3-dimensional volume of interest; AD, axial diffusivity; ADC, apparent diffusion coefficient; aMF, alternative motor fibers; CC, corpus callosum; CP, cerebral peduncle; CR, corona radiata; CS, centrum semiovale; CST, corticospinal tract; d-, difference between ipsilesional and contralesional; DTI, diffusion tensor imaging; FA, fractional anisotropy; i-, ipsilesional; MC, motor cortex; MD, medial diffusivity; PLIC, posterior limb of internal capsule; PMC, premotor cortex; PT, pyramidal tract; r-, ratio between ipsilesional and contralesional; RD, radial diffusivity.

TMS Biomarker Type

A number of TMS studies have shown that the presence of an MEP when stimulating the upper or lower extremity muscle representation areas of ipsilesional primary motor cortex (M1) is a good indicator of a significant motor recovery. In most cases, a TMS response in a specific arm/hand or leg/foot muscle was recorded as binary data (ie, absent or present).² Although the presence of MEP of upper or lower extremity is a crucial predictor of motor recovery, this variable is insufficient as a predictor alone in the model. To improve the accuracy of the model, the TMS biomarker should be incorporated with other neurologic and/or clinical biomarkers.⁵

sMRI Biomarker Type

Conventional sMRI studies usually used lesion location and volume information as predictors. Specifically, CST-lesion overlap volume (CST-lesion load) was calculated to quantify how much CST is damaged due to stroke.²⁸ In this review, we carefully used the term “CST structural integrity” separating it from “CST microstructural characteristics.” In previous literature, there was no distinction between “CST structural integrity” and “CST structural characteristics.” Investigators refer to DTI-derived metrics, such as FA and mean diffusivity (MD) of CST as “CST structural integrity,” but these DTI-derived metrics represent the water molecules’ diffusion directions and patterns along the axon.^24,29 Thus, using “structural integrity” for these DTI-derived variables lacks precision. We refer to “CST structural integrity” as the amount of damage to CST, which represents the overlap volume between the stroke lesion and the CST.

fMRI Biomarker Type

In functional imaging studies, the laterality index of ipsilesional M1 and functional connectivity between bilateral M1s during ipsilateral motor task performance were the most common predictor variables. Furthermore, recent studies have shown that functional connectivity among sensorimotor regions after stroke in resting-state can be a significant biomarker of brain functional impairment.^30-32

Combination Type

In studies using multiple neurological biomarkers, a combination of DTI and conventional sMRI biomarkers was the most common case. Further, a combination of DTI and TMS biomarkers was the next most frequently used. All combinations of biomarkers from included studies are listed in Table 14.

Table 14.

Summary of Combinations of Neurological Biomarkers.

Reference	Combinations of Neurological Biomarkers
Burke²⁸	DTI + fMRI
Dawes⁵⁷	DTI + sMRI
Granziera⁹	DTI + sMRI
Holtmannspötter⁶⁷	DTI + sMRI
Jang⁷	DTI + TMS
Lai⁸⁰	DTI + sMRI
Marumoto⁸⁸	DTI + sMRI
Nouri⁹⁰	sMRI + TMS
Nuutinen⁹¹	DWI + PWI + CT
Puig³⁷	DTI + sMRI
Quinlan³⁵	DTI + sMRI + fMRI
Rehme⁹⁵	fMRI + sMRI
Rosso⁹⁷	DTI + sMRI
Stinear²	DTI + TMS
Stinear¹³	DTI + TMS + fMRI
Tao⁹⁹	DTI + sMRI
Timmerhuis¹⁰⁰	TMS + sMRI
Wöhrle¹⁰³	DWI + PWI + TMS

Abbreviations: CT, computed tomography; DTI, diffusion tensor imaging; DWI, diffusion weighted imaging; fMRI, function magnetic resonance imaging; PWI, perfusion weighted imaging; sMRI, conventional structural magnetic resonance imaging (including T1-weighted and T2-weighted images); TMS, transcranial magnetic stimulation.

Dependent Variables (Clinical Endpoints) of Prediction Models

The most commonly employed clinical outcomes were the Rankin Scale (RS or modified RS), NIH Stroke Scale (or modified NIHSS) motor portion, and Barthel Index (BI). Although these clinical outcome measures have been proven to be highly reliable and valid,³³ they lack specificity for motor impairment or performance in individuals poststroke.³⁴ Furthermore, these low-level categorical scales lack sensitivity and resolution for detecting motor recovery.³³

There were several other clinical motor outcome variables that were utilized in reviewed studies, such as the Fugl-Meyer Assessment (FMA), Wolf Motor Function Test (WMFT) time score, Motricity Index (MI), Action Research Arm Test (ARAT), Nine-Hole Peg Test (NHPT), Grip force assessment, and walking performance measures. These measures represent the “Body Structure” and/or “Activity” levels of the International Classification of Functioning (ICF) and are more specific to motor impairment/performance.¹¹ As such, these motor-specific measures are more sensitive and may be more appropriate than other more generic and broad-based clinical outcome measures when the goal is to capture changes in motor behavior.

Recent studies have utilized a composite measure statistically derived from multiple clinical outcome measures, as a single outcome measure cannot capture all dimensions of motor recovery.³³ Quinlan et al³⁵ utilized principal component analysis (PCA) of 2 different motor outcome scores to specify changes in motor behavior. Incorporating several different motor outcome measures will improve the accuracy of model estimation by reducing the measurement errors in dependent variables. Although, using these data reduction methods, including PCA, is attractive from a statistical standpoint, the meaningfulness of the derived composite measure is not that transparent. Therefore, researchers should consider the pros and cons when composite measures of motor recovery are used as dependent variables of prediction models pertaining to changes in motor behavior.

Difference in Prediction Models Among the Acute, Subacute, and the Chronic Phases of Stroke

Most participants in the included studies were in the acute stroke phase rather than subacute or chronic phase (~56%). It has been suggested that prediction of motor recovery in the early phase of stroke may play an important role in tailoring neurorehabilitation therapies for each individual.^36-38 The initial assessment of brain impairment within 1 week after stroke including lesion volume or location was the most common predictor of motor recovery in previous studies. This might be because of the predominance of retrospective prognostic studies using diagnostic structural MRI. Furthermore, TMS was also frequently used to predict motor recovery during the acute phase. Researchers have reported that the presence of an MEP for hand muscles on the affected side during the acute phase is a strong predictor of upper extremity motor recovery.^2,13,39 DTI biomarkers were also frequently used as a predictor in the acute phase, but this is a controversial topic. There is evidence that DTI measures of CST cannot capture the structural impairment within 2 weeks after stroke.⁴⁰ This might be associated with the time course of Wallerian degeneration of white matter fibers after stroke.²⁷ Puig and colleagues⁴⁰ also showed that the FA ratio between ipsi- and contralesional CSTs acquired at admission and at day 3 after stroke is not a significant predictor of motor recovery, while these metrics acquired at day 30 is a strong predictor of motor recovery at 3 months. Furthermore, DTI biomarkers were predominantly used for the prediction model in the subacute phase. This suggests that a DTI measure of sensorimotor pathways taken during the subacute phase would be a viable predictor for motor recovery, while those captured during the acute phase would not.

Prediction of motor recovery at the chronic stroke phase may be much more complex. Individuals at the chronic stage may be near to their motor recovery potential,^41,42 and there are likely a number of additional secondary factors that can influence motor recovery, including psychosocial factors,⁴³ biomechanical factors,⁴⁴ motor learning,⁴⁵ and changes in brain structural and/or functional connectivity.⁴⁶ As such, it makes sense that prediction of motor recovery during this later phase should include multiple neurologic and clinical biomarkers to account for these additional secondary factors. A number of studies of individuals in the chronic stage after stroke have developed prediction models for motor improvement using neurological biomarkers.⁴⁷ Prediction models for motor recovery at the chronic stage are focused on whether the individual can benefit from specific treatments, such as transcranial direct current stimulation (t-DCS),⁴⁸ constraint-induced movement therapy (CIMT),⁴⁹ or behavioral neurorehabilitation therapies including task-specific motor practice.¹³ However, we lack high-quality prospective longitudinal clinical trials to develop a multimodal prediction model for motor recovery in the chronic phase.⁵⁰

Limitations

Our findings are limited by the prevalent use of broad-based clinical endpoints that lack sensitivity, specificity, and resolution for motor recovery, particularly that along the restitution-substitution continuum.⁵¹

Furthermore, several methodological weaknesses limited the impact of studies using functional MRI, conventional structural MRI, or TMS biomarkers. The reader is cautioned, however, that the low prevalence of methodologically robust prognostic studies that used these biomarkers as opposed to the more prevalent ones may have skewed our results.

Another limitation is that the evidence methodological evaluation tool used here is likely outdated and in need of revision. As the evidence evaluation tool was developed in 1990 when the imaging technology was in its infancy, several criteria may be inappropriate for recent developments in neuroimaging/neurophysiologic methods. Furthermore, although a number of items are critical to determining methodological quality of the evidence, all items carry the same weight. We attempted to minimize this limitation by redefining the criteria for several items; however, this is likely not a complete fix. The development of newer methodological quality evaluation tools would likely improve the systematic review process in this area.

Future Research

To improve methodological quality, we recommend that future studies: (a) perform cross-validation of the model, (b) consider the MCID of motor recovery outcome measures, and (c) recruit a large enough sample to provide sufficient statistical power. Furthermore, future studies that employ more sensitive and specific clinical endpoints coupled with valid neurological biomarkers are more likely to advance our understanding of the motor recovery process after stroke. Such an approach may lead to the development of more accurate prediction models than that achieved with the more traditional broad-based clinical endpoints that have dominated the literature, thus far.

Conclusion

Heterogeneity of poststroke brain pathology and motor impairment is a considerable challenge for the development of accurate prediction models. Accurate prediction of recovery is critical for determining the best neurorehabilitation protocol that will promote motor recovery and maximize meaningful outcomes. This focused systematic review found that conventional structural MRI and combination biomarker types possess the most methodologically robust evidence to be used for predicting motor recovery after stroke. Furthermore, it is not surprising that prediction models that used neurological biomarkers along with clinical measures (eg, Fugl-Meyer score, age, or chronicity) were more accurate than models that used neurological biomarkers alone.

Footnotes

Appendix

Supplementary material for this article is available on the Neurorehabilitation & Neural Repair website at

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

References

Bernhardt

Borschmann

Boyd

. Moving rehabilitation research forward: developing consensus statements for rehabilitation and recovery research. Int J Stroke. 2016;11:454-458.

Stinear

Barber

Petoe

Anwar

Byblow

. The PREP algorithm predicts potential for upper limb recovery after stroke. Brain. 2012;135:2527-2535.

Borich

Brown

Boyd

. Motor skill learning is associated with diffusion characteristics of white matter in individuals with chronic stroke. J Neurol Phys Ther. 2014;38:151-160.

Cramer

Parrish

Levy

. Predicting functional gains in a stroke trial. Stroke. 2007;38:2108-2114.

Feys

Van Hees

Bruyninckx

Mercelis

De Weerdt

. Value of somatosensory and motor evoked potentials in predicting arm recovery after a stroke. J Neurol Neurosurg Psychiatry. 2000;68:323-331.

Heiss

W-D

Kidwell

. Imaging for prediction of functional outcome and assessment of recovery in ischemic stroke. Stroke. 2014;45:1195-1201.

Jang

Ahn

Sakong

. Comparison of TMS and DTT for predicting motor outcome in intracerebral hemorrhage. J Neurol Sci. 2010;290:107-111.

Schiemanck

Kwakkel

Post

Prevo

. Predictive value of ischemic lesion volume assessed with magnetic resonance imaging for neurological deficits and functional outcome poststroke: a critical review of the literature. Neurorehabil Neural Repair. 2006;20:492-502.

Granziera

Daducci

Meskaldji

. A new early and automated MRI-based predictor of motor improvement after stroke. Neurology. 2012;79:39-46.

10.

Stinear

Byblow

Ward

. An update on predicting motor recovery after stroke. Ann Phys Rehabil Med. 2014;57:489-498.

11.

Chen

S-Y

Winstein

. A systematic review of voluntary arm recovery in hemiparetic stroke: critical predictors for meaningful outcomes using the international classification of functioning, disability, and health. J Neurol Phys Ther. 2009;33:2-13.

12.

Kwakkel

Kollen

. Predicting activities after stroke: What is clinically relevant? Int J Stroke. 2013;8:25-32.

13.

Stinear

Barber

Smale

Coxon

Fleming

Byblow

. Functional potential in chronic stroke patients depends on corticospinal tract integrity. Brain. 2007;130:170-180.

14.

Hier

Edelstein

. Deriving clinical prediction rules from stroke outcome research. Stroke. 1991;22:1431-1436.

15.

Kwakkel

Wagenaar

Kollen

Lankhorst

. Predicting disability in stroke—a critical review of the literature. Age Ageing. 1996;25:479-489.

16.

Symposium recommendations for methodology in stroke outcome research. Task Force on Stroke Impairment, Task Force on Stroke Disability, and Task Force on Stroke Handicap. Stroke. 1990;21:II68-II73.

17.

Gresham

. Stroke outcome research. Stroke. 1986;17:358-360.

18.

Atkins

Best

Briss

. Grading quality of evidence and strength of recommendations. BMJ. 2004;328:1490.

19.

Cohen

West

Aiken

. Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences. Neew York, NY: Routledge; 2013.

20.

Soper

. Post hoc statistical power calculator for multiple regression [Software]. 2015. http://www.danielsoper.com/statcalc/calculator.aspx?id=9. Accessed July 19, 2016.

21.

Kirschen

O’Higgins

Lee

. The Royal London Space Planning: an integration of space analysis and treatment planning: part I. Assessing the space required to meet treatment objectives. Am J Orthod Dentofacial Orthop. 2000;118:448-455.

22.

Weimar

König

Kraywinkel

Ziegler

Diener

, German Stroke Study Collaboration. Age and National Institutes of Health Stroke Scale Score within 6 hours after onset are accurate predictors of outcome after cerebral ischemia: development and external validation of prognostic models. Stroke. 2004;35:158-162.

23.

Lang

Edwards

Birkenmeier

Dromerick

. Estimating minimal clinically important differences of upper-extremity measures early after stroke. Arch Phys Med Rehabil. 2008;89:1693-1700.

24.

Soares

Marques

Alves

Sousa

. A hitchhiker’s guide to diffusion tensor imaging. Front Neurosci. 2013;7:31.

25.

Lindenberg

Renga

Zhu

Betzler

Alsop

Schlaug

. Structural integrity of corticospinal motor fibers predicts motor impairment in chronic stroke. Neurology. 2010;74:280-287.

26.

Scholz

Klein

Behrens

Johansen-Berg

. Training induces changes in white-matter architecture. Nat Neurosci. 2009;12:1370-1371.

27.

Puig

Pedraza

Blasco

. 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;31:1324-1330.

28.

Burke

Dobkin

Noser

Enney

Cramer

. Predictors and biomarkers of treatment gains in a clinical stroke trial targeting the lower extremity. Stroke. 2014;45:2379-2384.

29.

Sterr

Dean

PJ a.

Szameitat

a. J

Conforto

a. B

Shen

. Corticospinal tract integrity and lesion volume play different roles in chronic hemiparesis and its improvement through motor practice. Neurorehabil Neural Repair. 2014;28:335-343.

30.

Várkuti

Guan

Pan

. Resting state changes in functional connectivity correlate with movement recovery for BCI and robot-assisted upper-extremity training after stroke. Neurorehabil Neural Repair. 2013;27:53-62.

31.

Rehme

Eickhoff

Wang

Fink

Grefkes

. Dynamic causal modeling of cortical activity from the acute to the chronic stage after stroke. Neuroimage. 2011;55:1147-1158.

32.

Jung

T-D

Kim

J-Y

Seo

J-H

. Combined information from resting-state functional connectivity and passive movements with functional magnetic resonance imaging differentiates fast late-onset motor recovery from progressive recovery in hemiplegic stroke patients: a pilot study. J Rehabil Med. 2013;45:546-552.

33.

Kasner

. Clinical interpretation and use of stroke scales. Lancet Neurol. 2006;5:603-612.

34.

Bonita

Beaglehole

. Recovery of motor function after stroke. Stroke. 1988;19:1497-1500.

35.

Quinlan

Dodakian

See

. Neural function, injury, and stroke subtype predict treatment gains after stroke. Ann Neurol. 2015;77:132-145.

36.

Stinear

. Prediction of recovery of motor function after stroke. Lancet Neurol. 2010;9:1228-1232.

37.

Puig

Pedraza

Blasco

. 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;32:857-863.

38.

Groisser

Copen

Singhal

Hirai

Schaechter

. Corticospinal tract diffusion abnormalities early after stroke predict motor outcome. Neurorehabil Neural Repair. 2014;28:751-760. Erratum in: Neurorehabil Neural Repair. 2015;29:296.

39.

Arac

Sagduyu

Binai

Ertekin

. Prognostic value of transcranial magnetic stimulation in acute stroke. Stroke. 1994;25:2183-2186. Erratum in: Stroke. 1995;26:336.

40.

Puig

Blasco

Daunis-I-Estadella

. Decreased corticospinal tract fractional anisotropy predicts long-term motor outcome after stroke. Stroke. 2013;44:2016-2018.

41.

van Kordelaar

van Wegen

Kwakkel

. Impact of time on quality of motor control of the paretic upper limb after stroke. Arch Phys Med Rehabil. 2014;95:338-344.

42.

Winters

van Wegen

Daffertshofer

Kwakkel

. Generalizability of the proportional recovery model for the upper extremity after an ischemic stroke. Neurorehabil Neural Repair. 2015;29:614-622.

43.

Winstein

Lewthwaite

Blanton

Wolf

Wishart

. Infusing motor learning research into neurorehabilitation practice: a historical perspective with case exemplar from the accelerated skill acquisition program. J Neurol Phys Ther. 2014;38:190-200.

44.

Gao

Grant

Roth

Zhang

. Changes in passive mechanical properties of the gastrocnemius muscle at the muscle fascicle and joint levels in stroke survivors. Arch Phys Med Rehabil. 2009;90:819-826.

45.

Winstein

Merians

Sullivan

. Motor learning after unilateral brain damage. Neuropsychologia. 1999;37:975-987.

46.

Crofts

Higham

Bosnell

. Network analysis detects changes in the contralesional hemisphere following stroke. Neuroimage. 2011;54:161-169.

47.

Lindenberg

Zhu

Rüber

Schlaug

. Predicting functional motor potential in chronic stroke patients using diffusion tensor imaging. Hum Brain Mapp. 2012;33:1040-1051.

48.

O’Shea

Boudrias

Stagg

. Predicting behavioural response to TDCS in chronic motor stroke. Neuroimage. 2014;85:924-933.

49.

Rickards

Sterling

Taub

. Diffusion tensor imaging study of the response to constraint-induced movement therapy of children with hemiparetic cerebral palsy and adults with chronic stroke. Arch Phys Med Rehabil. 2014;95:506-514.e1.

50.

Stinear

Byblow

. Predicting and accelerating motor recovery after stroke. Curr Opin Neurol. 2014;27:624-630.

51.

Levin

Kleim

Wolf

. What do motor “recovery” and “compensation” mean in patients following stroke? Neurorehabil Neural Repair. 2009;23:313-319.

52.

Kelly

Furie

Shafqat

Rallis

Chang

Stein

. Functional recovery following rehabilitation after hemorrhagic and ischemic stroke. Arch Phys Med Rehabil. 2003;84:968-972.

53.

Chang

Chun

. Prediction of lower limb motor outcomes based on transcranial magnetic stimulation findings in patients with an infarct of the anterior cerebral artery. Somatosens Mot Res. 2015;32:249-253.

54.

Cho

S-H

Kim

D-S

Kim

Y-H

Lee

C-H

Jang

. 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. 2007;426:123-127.

55.

Cho

S-H

Kim

Choi

. Motor outcome according to diffusion tensor tractography findings in the early stage of intracerebral hemorrhage. Neurosci Lett. 2007;421:142-146.

56.

Coutts

O’Reilly

Hill

. Computed tomography and computed tomography angiography findings predict functional impairment in patients with minor stroke and transient ischaemic attack. Int J Stroke. 2009;4:448-453.

57.

Dawes

Enzinger

Johansen-Berg

. Walking performance and its recovery in chronic stroke in relation to extent of lesion overlap with the descending motor tract. Exp Brain Res. 2008;186:325-333.

58.

Di Lazzaro

Profice

Pilato

. Motor cortex plasticity predicts recovery in acute stroke. Cereb Cortex. 2010;20:1523-1528.

59.

Dong

Dobkin

Cen

Winstein

. Motor cortex activation during treatment may predict therapeutic gains in paretic hand function after stroke. Stroke. 2006;37:1552-1555.

60.

Escudero

J V.

Sancho

Bautista

Escudero

Lopez-Trigo

. Prognostic value of motor evoked potential obtained by transcranial magnetic brain stimulation in motor function recovery in patients with acute ischemic stroke. Stroke. 1998;29:1854-1859.

61.

Feng

Wang

Chhatbar

. Corticospinal tract lesion load: An imaging biomarker for stroke motor outcomes. Ann Neurol. 2015;78:860-870.

62.

Feydy

Carlier

Roby-Brami

. Longitudinal study of motor recovery after stroke: recruitment and focusing of brain activation. Stroke. 2002;33:1610-1617.

63.

Grässel

Ringer

Fitzek

. Wallerian degeneration of pyramidal tract after paramedian pons infarct. Cerebrovasc Dis. 2010;30:380-388.

64.

Hand

Wardlaw

Rivers

. MR diffusion-weighted imaging and outcome prediction after ischemic stroke. Neurology. 2006;66:1159-1163.

65.

Hendricks

Hageman

van Limbeek

. Prediction of recovery from upper extremity paralysis after stroke by measuring evoked potentials. Scand J Rehabil Med. 1997;29:155-159.

66.

Hendricks

Pasman

van Limbeek

Zwarts

. Motor evoked potentials of the lower extremity in predicting motor recovery and ambulation after stroke: a cohort study. Arch Phys Med Rehabil. 2003;84:1373-1379.

67.

Holtmannspötter

Peters

Opherk

. Diffusion magnetic resonance histograms as a surrogate marker and predictor of disease progression in CADASIL: a two-year follow-up study. Stroke. 2005;36:2559-2565.

68.

Jang

Cho

S-H

Kim

Y-H

. Diffusion anisotrophy in the early stages of stroke can predict motor outcome. Restor Neurol Neurosci. 2005;23:11-17.

69.

Jang

Kim

Y-H

Chang

Han

Byun

Chang

. The predictive value of cortical activation by passive movement for motor recovery in stroke patients. Restor Neurol Neurosci. 2004;22:59-63.

70.

Kang

H-J

Stewart

Park

M-S

. White matter hyperintensities and functional outcomes at 2 weeks and 1 year after stroke. Cerebrovasc Dis. 2013;35:138-145.

71.

Karibe

Shimizu

Tominaga

Koshu

Yoshimoto

. Diffusion-weighted magnetic resonance imaging in the early evaluation of corticospinal tract injury to predict functional motor outcome in patients with deep intracerebral hemorrhage. J Neurosurg. 2000;92:58-63.

72.

Kim

Y-H

Kim

Park

Lee

Chang

. Prediction of motor recovery using diffusion tensor tractography in supratentorial stroke patients with severe motor involvement. Ann Rehabil Med. 2015;39:570-576.

73.

Koh

C-L

Pan

S-L

Jeng

J-S

. Predicting recovery of voluntary upper extremity movement in subacute stroke patients with severe upper extremity paresis. PLoS One. 2015;10:e0126857.

74.

Koski

Mernar

Dobkin

. Immediate and long-term changes in corticomotor output in response to rehabilitation: correlation with functional improvements in chronic stroke. Neurorehabil Neural Repair. 2004;18:230-249.

75.

Koyama

Tsuji

Miyake

Ohmura

Domen

. Motor outcome for patients with acute intracerebral hemorrhage predicted using diffusion tensor imaging: an application of ordinal logistic modeling. J Stroke Cerebrovasc Dis. 2012;21:704-711.

76.

Koyama

Tsuji

Nishimura

Miyake

Ohmura

Domen

. 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;22:72-79.

77.

Kusano

Seguchi

Horiuchi

. Prediction of functional outcome in acute cerebral hemorrhage using diffusion tensor imaging at 3T: a prospective study. AJNR Am J Neuroradiol. 2009;30:1561-1565.

78.

Kuzu

Inoue

Kanbara

. Prediction of motor function outcome after intracerebral hemorrhage using fractional anisotropy calculated from diffusion tensor imaging. Cerebrovasc Dis. 2012;33:566-573.

79.

Kwon

Jeoung

Lee

. Predictability of motor outcome according to the time of diffusion tensor imaging in patients with cerebral infarct. Neuroradiology. 2012;54:691-697.

80.

Lai

Zhang

Liu

. White matter tractography by diffusion tensor imaging plays an important role in prognosis estimation of acute lacunar infarctions. Br J Radiol. 2007;80:782-789.

81.

Lai

Wang

Tsai

. Corticospinal integrity and motor impairment predict outcomes after excitatory repetitive transcranial magnetic stimulation : a preliminary study. Arch Phys Med Rehabil. 2015;96:69-75.

82.

Liggins

Yoo

Mishra

.; DEFUSE 2 Investigators. A score based on age and DWI volume predicts poor outcome following endovascular treatment for acute ischemic stroke. Int J Stroke. 2015;10:705-709.

83.

Liu

Tian

Qiu

. Correlation analysis of quantitative diffusion parameters in ipsilateral cerebral peduncle during Wallerian degeneration with motor function outcome after cerebral ischemic stroke. J Neuroimaging. 2012;22:255-260.

84.

Loubinoux

Carel

Pariente

. Correlation between cerebral reorganization and motor recovery after subcortical infarcts. Neuroimage. 2003;20:2166-2180.

85.

Gao

. Effect of baseline magnetic resonance imaging (MRI) apparent diffusion coefficient lesion volume on functional outcome in ischemic stroke. Neurol Res. 2011;33:494-502.

86.

Liu

Zhou

. Longitudinal study of diffusion tensor imaging properties of affected cortical spinal tracts in acute and chronic hemorrhagic stroke. J Clin Neurosci. 2014;21:1388-1392.

87.

Maeshima

Osawa

Nishio

Hirano

Kigawa

Takeda

. Diffusion tensor MR imaging of the pyramidal tract can predict the need for orthosis in hemiplegic patients with hemorrhagic stroke. Neurol Sci. 2013;34:1765-1770.

88.

Marumoto

Koyama

Hosomi

. Diffusion tensor imaging predicts the outcome of constraint-induced movement therapy in chronic infarction patients with hemiplegia: a pilot study. Restor Neurol Neurosci. 2013;31:387-396.

89.

Nascimbeni

Gaffuri

Imazio

. Motor evoked potentials: prognostic value in motor recovery after stroke. Funct Neurol. 2006;21:199-203.

90.

Nouri

Cramer

. Anatomy and physiology predict response to motor cortex stimulation after stroke. Neurology. 2011;77:1076-1083.

91.

Nuutinen

Liu

Laakso

. Assessing the outcome of stroke: a comparison between MRI and clinical stroke scales. Acta Neurol Scand. 2006;113:100-107.

92.

Piron

Piccione

Tonin

Dam

. Clinical correlation between motor evoked potentials and gait recovery in poststroke patients. Arch Phys Med Rehabil. 2005;86:1874-1878.

93.

Qiu

Zhang

. Preliminary application of pyramidal tractography in evaluating prognosis of patients with hypertensive intracerebral hemorrhage. Acta Neurochir Suppl. 2008;105:165-170.

94.

Rapisarda

Bastings

de Noordhout

Pennisi

Delwaide

. Can motor recovery in stroke patients be predicted by early transcranial magnetic stimulation? Stroke. 1996;27:2191-2196.

95.

Rehme

Volz

Feis

Eickhoff

Fink

Grefkes

. Individual prediction of chronic motor outcome in the acute post-stroke stage: behavioral parameters versus functional imaging. Hum Brain Mapp. 2015;36:4553-4565.

96.

Rickards

Taub

Sterling

. Brain parenchymal fraction predicts motor improvement following intensive task-oriented motor rehabilitation for chronic stroke. Restor Neurol Neurosci. 2012;30:355-361.

97.

Rosso

Colliot

Pires

. Early ADC changes in motor structures predict outcome of acute stroke better than lesion volume. J Neuroradiol. 2010;38:105-112.

98.

Schiemanck

Kwakkel

Post

Kappelle

Prevo

. Predicting long-term independency in activities of daily living after middle cerebral artery stroke: does information from MRI have added predictive value compared with clinical information? Stroke. 2006;37:1050-1054.

99.

Tao

W-D

Wang

Schlaug

Liu

Selim

. A comparative study of fractional anisotropy measures and ICH score in predicting functional outcomes after intracerebral hemorrhage. Neurocrit Care. 2014;21:417-425.

100.

Timmerhuis

Hageman

Oosterloo

Rozeboom

. The prognostic value of cortical magnetic stimulation in acute middle cerebral artery infarction compared to other parameters. Clin Neurol Neurosurg. 1996;98:231-236.

101.

van Kuijk

Pasman

Hendricks

Zwarts

Geurts

. 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.

102.

Wang

D-M

Liu

J-R

H-Y

. Diffusion tensor imaging predicts long-term motor functional outcome in patients with acute supratentorial intracranial hemorrhage. Cerebrovasc Dis. 2012;34:199-205.

103.

Wöhrle

Behrens

Mielke

Hennerici

. Early motor evoked potentials in acute stroke: adjunctive measure to MRI for assessment of prognosis in acute stroke within 6 hours. Cerebrovasc Dis. 2004;18:130-134.

104.

Yoshioka

Horikoshi

Aoki

. Diffusion tensor tractography predicts motor functional outcome in patients with spontaneous intracerebral hemorrhage. Neurosurgery. 2008;62:97-103.

105.

Zhu

Zhang

. A longitudinal diffusion tensor imaging study on Wallerian degeneration of corticospinal tract after motor pathway stroke. Neuroimage. 2009;47:451-458.

106.

Zarahn

Alon

Ryan

. Prediction of motor recovery using initial impairment and fMRI 48 h poststroke. Cereb Cortex. 2011;21:2712-2721.