PRECISION NEUROREHABILITATION IN CHRONIC STROKE: BIOMARKER-GUIDED, AI-OPTIMIZED DOSING OF TASK-SPECIFIC TRAINING TO ENHANCE CORTICOSPINAL NEUROPLASTICITY

Abstract

Background: Chronic stroke recovery plateaus reflect heterogeneous neurobiology and fixed-dose rehabilitation. Precision approaches integrating biomarkers with adaptive dosing may optimize corticospinal neuroplasticity. Objective: To evaluate whether biomarker-guided, AI-optimized task-specific training (TST) enhances corticospinal tract (CST) neuroplasticity and motor function versus fixed-dose TST in chronic stroke. Methods: In a multicenter, assessor-blinded RCT, 88 chronic stroke survivors (≥6 months; FMA-UE 10–50) were randomized to AI-closed-loop dosing (biomarker-initialized) or standardized TST for 12 weeks. Primary outcomes: TMS-evoked MEP amplitude and CST fractional anisotropy. Secondary outcomes included clinical motor scales and kinematics. Linear mixed-effects and mediation analyses were applied. Results: AI-optimized TST yielded superior MEP amplitude (F(1,84) = 14.36, p < .001) and CST FA (p = .001) gains, sustained at 24 weeks. Functional improvement was greater (FMA-UE: +14.4 vs. +6.1, p < .001), with 42% of gains mediated by CST neuroplasticity. AI dosing achieved 92.8% adherence without increased adverse events. Conclusion: Biomarker-initialized, AI-optimized TST safely drives dose-dependent corticospinal remodeling, translating neural adaptation into functional recovery. This closed-loop precision framework warrants clinical adoption.