Abstracts

Rationale: Cortical stimulation mapping (CSM) remains the gold standard methodology for the localization of eloquent cortex. The resulting language map is often used in conjunction with structural imaging and intracranial electrophysiology to inform the clinical approach. Here, we combine these functional and structural measures in a large patient population to understand the architecture of spatially distributed substrates supporting language function. Methods: We collected data in 214 patients undergoing language mapping (awake craniotomy, n=63, subdural grids, n=49; stereotactic depths, n=102) with CSM (1-10mA, 50Hz, 2s), diffusion tensor imaging (DTI), and/or intracranial electrophysiology. Language function was evaluated with a battery of tasks including visual picture naming and auditory naming to definition. Stimulation-induced depolarization and electrode recording zones were transformed onto the pial surface with a current spread model to generate subject-specific functional map. These were then co-registered with DTI. Gamma (60-120 Hz) power during task performance was strongly predictive of functional classification by CSM. In several regions of interest, we generated distinct volumetric density representations of white matter fibers underlying either positive or negative stimulation sites, as well as either active or quiescent electrodes. Tracts were distilled from the sets of positive and negative fibers using an unsupervised clustering algorithm – DB-SCAN. The resulting subject-specific maps (surface-based) and tracts (volumetric) were projected into a standard atlas space with a nonlinear combined surface-volume transform that maximizes surface-landmark similarity and minimizes volumetric distortion. Results: CSM at the group-level revealed five regions which consistently disrupted both auditory and visual naming function: middle fusiform gyrus, inferior frontal gyrus, dorsomedial prefrontal cortex, superior temporal gyrus, and posterior middle temporal gyrus. These regions were also identified using electrophysiology and are listed in their temporal sequence of engagement. DTI maps in each of these regions of interest revealed strong intra-network connectivity for positive stimulation sites, but diffuse and inconsistent cortical connectivity for negative stimulation sites. Conclusions: This analysis, integrating three essential surgical planning tools, constitutes a significance advance in large-scale multimodal population-level maps of human language. Funding: 1F30DC017083