EMG-derived muscle synergy, as a representation of neuromotor modules utilized for motor control, has been proposed as a biomarker for stroke rehabilitation. Here, we evaluate the utility of muscle synergies for assessing motor function and predicting post-intervention motor outcome in a stroke rehabilitation clinical trial. Subacute stroke survivors (n = 59) received month-long acupuncture (Acu), sham acupuncture (ShamAcu) or no acupuncture (NoAcu) as adjunctive rehabilitative intervention alongside standard physiotherapy. Clinical scores and EMGs (14 muscles, eight motor tasks) were collected from the stroke-affected upper limb before and after intervention. We then extracted muscle synergies from EMGs using non-negative matrix factorization and designed 12 muscle synergy indices (MSIs) to summarize different aspects of post-stroke synergy features. All MSIs correlated with multiple clinical scores, suggesting that our indices could potentially serve as biomarkers for post-stroke motor functional assessments. While the intervention groups did not differ in their pre-to-post differences in the clinical scores, the inclusion of MSIs into analysis revealed that on average Acu promoted more recovery of synergy features than ShamAcu and NoAcu, though not all subjects in the group were Acu responders. We then built regression models using pre-intervention MSIs and clinical variables to predict the outcomes of Acu and NoAcu and showed by a preliminary retrospective simulation of patient stratification that MSI-based predictions could have led to better post-intervention motor improvement. Overall, we demonstrate that muscle synergies can potentially clarify the effects of interventions and assist in motor assessment, outcome prediction, and treatment selection.
Motor rehabilitation after stroke is challenging due to the heterogeneity of stroke presentations. Maximizing recovery hinges on suitable personalization of rehabilitation, which depends on reliable motor assessments and predictions of the subjects’ responses to different interventions using biomarkers of brain impairment. Beyond the previously defined neuroimaging biomarkers, impairment-dependent motor patterns of stroke survivors during voluntary movement are alternatives that potentially offer accurate and precise predictions. Specifically, muscle synergies identified from multi-muscle electromyographic signals (EMG), as neuromotor modules employed by CNS for muscle coordination, have been previously used to evaluate upper limb functions post-stroke in small-to-moderate-sized cohorts. While these initial results appear promising, how muscle synergies should be most profitably used for clinical assessments and whether they predict post-rehabilitation responses remain unexplored. Here, we evaluate the potential of muscle synergies in assessing upper limb motor functions and predicting outcomes from multiple rehabilitative options in a moderately large cohort of subacute stroke survivors (N=88, 55±35 days post-stroke) recruited for a randomized trial of acupuncture as an adjunctive rehabilitative intervention. Subjects (N=59) were randomly assigned to monthlong abdominal acupuncture (Acu), sham acupuncture (ShamAcu), or no acupuncture (NoAcu), alongside basic care. Four clinical scores and EMGs of the stroke-affected upper limb (14 muscles, 8 tasks) were collected before and after intervention. Muscle synergies were extracted from EMGs using factorization. For each subject, features of the synergies and their temporal activations were comprehensively summarized by 12 muscle synergy indexes (MSI). We first demonstrated cross-sectionally that our MSIs correlated significantly with all clinical scores, and thus could capture impairment-related synergy changes. Longitudinally, Acu was differentiated from ShamAcu and NoAcu in having clinical score improvements accompanied by the restorations of more MSIs. For each treatment group, we then built regression models that predict clinical scores’ realized recovery from pre-intervention MSIs and other variables. Model-predicted recovery correlated significantly with observed recovery (R 2 =0.53-0.70). To test the models’ utility in patient stratification, for every Acu and NoAcu subject we retrospectively identified the intervention option expected to yield greater recovery by comparing the predicted Acu and NoAcu outcomes. Subjects who indeed received the model-assigned intervention showed more realized recovery in Fugl-Meyer Assessment (section A) than those who received incorrectly assigned intervention (p=0.013). Overall, our findings suggest that muscle synergies, when suitably summarized as MSIs, may clarify the intervention’s effects and assist in motor assessment, outcome prediction, and treatment selection. MSIs can be useful recovery biomarkers in future schemes of precision rehabilitation.