CNSC-97. ELECTROCORTICOGRAPHIC ANALYSIS AND DETECTION OF SPEECH NETWORKS IN GLIOMA-INFILTRATED CORTEX

Abstract Direct cortical stimulation (DCS) is the clinical gold standard for identifying the functional cortex in the human brain, essential for the safe removal of brain lesions. However, DCS can be time- and resource-intensive and yields only one datapoint on each trial. In contrast, subdural electrocorticography (ECoG) is multidimensional, yielding measures of activity across a cortical region. Further, it is unknown which electro-physiological features reliably predict DCS+ regions in patients with gliomas. Accordingly, defining the electro-physiological properties of DCS-positive cortical regions may facilitate the identification of critical language regions, thereby permitting safe glioma resections in communities without access. Leveraging a prospectively collected multicenter electrophysiologic dataset of DCS-positive language regions spatially matched with a subdural ECoG array, we analyzed regions identified as functionally critical (DCS+) versus functionally non-critical (DCS–) during intraoperative language mapping (81 patients). In IDH-mutant gliomas (36 patients; 39 DCS+, 576 DCS–), DCS+ regions exhibited significantly greater, more coordinated speech-related neural activity in the high gamma frequency range (70-150 Hz), indicating stronger task-related electrophysiological coordination at DCS+ sites. These differences persisted even when pair matching each DCS+ to the closest DCS– site (within the same patient/gyrus). Interestingly, these differences were not apparent in the IDH-wildtype cohort (45 patients; 44 DCS+, 479 DCS–). Further, neuronal high gamma time series from IDH-mutant DCS+ sites demonstrated enhanced encoding and decoding of linguistic and semantic features (using neuronal recordings to predict/classify word-level features). Again, this was not apparent in the IDH-wildtype cohort. Finally, we demonstrate that resting-state classifiers can accurately distinguish DCS+ from DCS– regions in IDH-mutant tumors (>88% accuracy, AUROC of 0.817, sensitivity of 0.80, a specificity of 0.917 in an external validation cohort). This capability has the potential to accelerate DCS mapping by guiding surgeons to priority regions, thereby improving surgical efficiency and patient outcomes. As the task-based and resting-state electrophysiologic distinctions were not observed in IDH-wildtype glioblastomas, this suggests a pathology-specific remodeling of cortical language networks following glioma infiltration.