Neuromechanical Predictors of Clinical Scores of Balance and Functional Mobility in Chronic Stroke Survivors - A Machine Learning Approach.

Clinical tests such as the Berg Balance Scale (BBS) and Timed Up and Go (TUG) are used to assess balance and functional mobility following stroke. These tests are subjective due to dependence on the assessor's judgment and the patient's effort, potentially affecting their sensitivity to early or subtle balance recovery. This study aimed to identify objective markers of BBS and TUG among corticomuscular coherence (CMC) and force platform center of pressure (COP) measures using machine learning in 18 chronic stroke patients and 15 age-matched healthy adults. Participants performed a continuous balance task on a sway-referenced force platform with simultaneous recording of EEG, EMG, and COP data. We used a two-stage machine learning approach: first, a binary classifier, such as XGBoost and Elastic Net models, reduced dimensionality while preserving interpretability, and selected features that differentiated between stroke and healthy controls; second, regression models used selected features to identify predictors of BBS and TUG. Tibialis anterior delta-band CMC and medio-lateral root mean square COP predicted BBS and TUG. These features capture the dynamic stability mechanisms shared across the two clinical tests of balance and functional mobility. Soleus theta-band CMC asymmetry index predicted BBS, whereas TUG was predicted by biceps femoris beta-band CMC asymmetry index and rectus femoris beta-band CMC. These muscle-specific measures highlighted the use of ankle and hip strategies for the control of balance during BBS and TUG tests, respectively. Our study provides objective neurophysiological and biomechanical markers that may be sensitive to subtle changes in balance following stroke.