Fig. 2: Direct electrical stimulation (DES) derived structural networks predict the severity of stroke-induced aphasia symptoms and can be used to model post-operative language recovery trajectories in glioma patients.

(Upper panel) We found that the volume of overlap between the stroke lesion and the volumetric normative maps for semantics and phonological processing significantly correlates with symptom severity—as assessed by the aphasia quotient score from the Western Aphasia Battery (revised version)—and yields better predictions than total lesion size. (lower panel) We compared DES-derived structural networks against state-of-the-art clinical and demographic variables (see main text) in predicting the longitudinal semantic and phonemic fluencies measured pre-, 1 week, and 1 month after surgery in glioma patients using a series of piecewise linear mixed models. In a head-to-head comparison, we found that adding the overlap between the tumor cavity and the normative structural maps to a model containing only a clinical control variable explained significantly more variance than adding the clinical control variable to a model containing only the overlap between the tumor cavity and the normative structural maps. For each comparison, the green bar recapitulates the increase in variance explained—expressed as effect size—by adding the overlap between the tumor cavity and the normative structural maps, the non-green bar illustrates the effect size increase achieved by adding the clinical control variable. We sorted the comparison by increasing magnitude of the difference in variance explained (dashed lines).