Multivariate machine learning-based language mapping in glioma patients based on lesion topography.

Abstract

Diffusive and progressive tumor infiltration within language-related areas of the brain induces functional reorganization. However, the macrostructural basis of subsequent language deficits is less clear. To address this issue, lesion topography data from 137 preoperative patients with left cerebral language-network gliomas (81 low-grade gliomas and 56 high-grade gliomas), were adopted for multivariate machine-learning-based lesion-language mapping analysis. We found that tumor location in the left posterior middle temporal gyrus-a bottleneck where both dorsal and ventral language pathways travel-predicted deficits of spontaneous speech (cluster size = 1356 mm3, false discovery rate corrected P < 0.05) and naming scores (cluster size = 1491 mm3, false discovery rate corrected P < 0.05) in the high-grade glioma group. In contrast, no significant lesion-language mapping results were observed in the low-grade glioma group, suggesting a large functional reorganization. These findings suggest that in patients with gliomas, the macrostructural plasticity mechanisms that modulate brain-behavior relationships depend on glioma grade.