The relationship between functional recovery and indirect disconnection measures in stroke

Background Normative atlas-based methods for estimating disconnection from brain lesions are increasingly used, but their predictive value for clinical recovery remains unclear. We tested whether these disconnectivity measures improve stroke outcome prediction beyond lesion characteristics and clinical factors. Methods We retrospectively included stroke patients with structural MRI or CT and 6- or 12-month Stroke Impact Scale (SIS) scores. For each patient, we computed lesion volume, location, and maps of structural and functional disconnection using normative atlases. We also calculated the proportion of known brain networks affected. SIS subitems underwent factorial analysis to extract components used as outcomes in regression models with bagging. Three models (lesion features, disconnection patterns at voxel-level, and atlas-based disconnection topology) were compared to a reference model (demographic/clinical variables) using bootstrapping and adjusted R 2 . Results Seventy-four patients (mean age 66.8±10) were analyzed. Three SIS components emerged: mobility, cognition, and emotion. Models using disconnection features explained little variance (R²<0.07) for all components. Functional topology moderately predicted mobility (R²=0.202), significantly outperforming the reference model (R²=0.165, p=0.007). Adding functional topology to clinical variables raised mobility prediction to 33%. Predictability for cognition and emotion remained low. Conclusions The SIS shows three components that can be predicted differently based on distinct lesion properties. Overall, the prediction of indirect disconnection measures was low for all three components when considering disconnection patterns. On the contrary, the predictive value of a functional atlas-based framework was higher for the mobility factor. The cognitive and emotion factors presented lower values than the mobility component.