Structural Path Lengths Predict Post-Stroke Upper Limb Functional Outcome: Activation Likelihood vs. Whole Brain...

This retrospective study aimed to compare the predictive utility of shortest structural path length (SSPL) features derived from a whole-brain atlas and a task-specific atlas for forecasting post-stroke upper-limb motor outcomes.

Data from 142 individuals with stroke were analysed. Baseline assessments included clinical measures (National Institutes of Health Stroke Scale [NIHSS], Shoulder Abduction and Finger Extension [SAFE] score, and Fugl-Meyer Assessment of the Upper Extremity [FMUE]), neurophysiological measures (motor evoked potential [MEP] status), and demographic variables. SSPL features were quantified using the Lesion Quantification Toolkit with connectivity defined by the HCP-842 atlas. Features were extracted from two primary parcellation frameworks: a whole-brain Schaefer 100 atlas and a task-specific, sensorimotor-activation likelihood estimation-based atlas (SMAA). Six regression models with nested resampling were trained to predict 12-week upper-limb outcomes, and feature contributions were assessed using Shapley Additive Explanations (SHAP).

Baseline clinical models consistently outperformed SSPL-based models. SMAA-derived SSPL features achieved predictive performance comparable to that of the whole brain atlas but did not provide substantial additional prognostic value when combined with baseline clinical measures.

Baseline clinical and neurophysiological measures, particularly MEP and motor impairment scales, remain the strongest predictors of post-stroke upper-limb recovery. Although task-specific parcellations may provide a more compact and interpretable representation of motor-network disruption than whole-brain approaches, their incremental predictive value beyond established clinical predictors was limited.