Predicting Motor Recovery After Stroke: Utility and Limits of Corticospinal Tract Biomarkers

The study compared conventional methods and machine learning in predicting motor recovery after stroke based on corticospinal tract damage in 127 patients with acute ischemic stroke followed for more than 3 months. All conventional corticospinal tract metrics significantly correlated with acute impairment and motor outcome, with metrics accounting for tract narrowing performing best. Machine learning outperformed conventional markers in predicting acute impairment and basic motor outcome, but failed to predict complex motor outcome. Predictive voxels clustered in the posterior limb of the internal capsule, with different corticospinal tract subregions associated with different types of motor impairment. Results indicate that the predictive value of biomarkers depends on the type of motor function and time after stroke.