Post-stroke outcome prediction based on lesion-derived features

Stroke-induced deficits result from both focal structural damage and widespread network disruption. This study investigated whether simulated measures of network disruption, derived from structural lesions, could predict functional impairments in stroke patients. We extracted four lesion-derived feature sets: lesion masks, probabilistic structural disconnection maps (pSDMs), structural and indirectly estimated functional connectivity strengths between brain regions, and topological properties of functional and structural brain networks to predict motor, executive, and processing speed deficits in 340 S patients, employing PCA-based ridge regression with leave-one-out cross validation. The findings revealed that both structural disconnection map patterns and lesion masks were strong predictors of functional deficits. Lesion masks exhibited superior predictive performance relative to unthresholded pSDMs. Furthermore, applying a probability threshold to the pSDMs- retaining only disconnections present in a sufficient proportion of healthy subjects- significantly improved predictive performance. For motor deficits, thresholded SDMs (tSDMs) with thresholds of 0.9 and 0.5 produced the highest R2 values, 0.95 for left motor deficits and 0.69 for right motor deficits, respectively. In the case of executive function and processing speed, the highest R2 values were 0.58 and 0.64, achieved with tSDM thresholds of 0.3 and 0.5, respectively. Connectomebased features exhibited lower R2 values, with structural connection strength alterations showing stronger associations with post-stroke scores compared to changes in functional connectivity. Nodal parameters (degree and clustering coefficient) had lower explanatory power than the SDM features and lesion masks. Our findings underscore the effectiveness of lesion masks and thresholded SDMs in predicting post-stroke deficits. This study contributes to the growing body of evidence supporting the reliability of simulated structural networks as a complementary approach to lesion patterns and structural disconnection in predicting post- stroke outcomes.