Network measures from the REWIRED simulation framework enhance prediction of post-stroke aphasia severity

Predicting post-stroke aphasia severity remains challenging, in part because language outcomes reflect not only focal cortical damage but also widespread disruption of structural and functional networks. Computational models of large-scale cortical dynamics offer a principled way to infer these network-level consequences from patient-specific lesions. Here, we present and evaluate REWIRED, a lesion-informed cortical dynamics framework designed to simulate individualized alterations in large-scale brain network organization after stroke. We first evaluated whether simulation-derived functional connectivity captured patient-specific variation in empirical functional connectivity beyond lesion burden and structural disconnection alone. We then developed a multiscale feature set combining lesion volume, lesion distribution patterns, probabilistic disconnectome metrics, and simulation-derived measures of functional connectivity and effective information flow (EIF). Finally, using a nested support vector regression (SVR) framework in a separate dataset, we tested whether simulation-derived features improve prediction of chronic aphasia severity, measured by the Western Aphasia Battery - Revised Aphasia Quotient (WAB-AQ), beyond lesion-distribution and structural-connectivity predictors. Simulation-derived functional connectivity significantly predicted empirical functional connectivity beyond local lesion burden and structural disconnection alone. With respect to WAB-AQ prediction, lesion-based (Set 1) and disconnectome-based (Set 2a) features alone yielded modest accuracy. Adding simulation-derived features (Set 2b) produced substantial gains, and the full feature set (Set 3) achieved the best performance (RMSE = 14.5; r = 0.83), reaching accuracy that is competitive with recent multimodal neuroimaging approaches, despite relying solely on lesion-distribution inputs. EIF measures were consistently selected as top predictors, indicating that disruptions in interregional communication patterns carry behaviorally relevant information not captured by structural features alone. These results support REWIRED as a framework for linking structural injury to distributed network dysfunction and behavioral outcomes. By integrating lesion information with large-scale cortical dynamics modeling, REWIRED provides a foundation for future individualized modeling of recovery and rehabilitation.