NIH-funded Project on Language Localization

Title

Narrative

This project provides a new model that accurately predicts language outcomes after individually-tailored epilepsy surgery. The prior project validated our language mapping method based on naming task-related neural responses on intracranial EEG (iEEG). Our machine learning model incorporating multiple iEEG frequency responses, including high gamma augmentation at 70-110 Hz, predicted patients developing postoperative language outcomes as defined by objective neuropsychological assessment. The accuracy of our prior model, primarily considering how much language cortices were preserved, was 0.80, which is significant but not optimal (e.g., 0.90). We hypothesize that the model performance will be optimized by considering both neural communications via white matter tracts between language cortices and the expected reduction of seizure-related burdens after surgery. In this project, our prediction model will newly incorporate the language-related white matter pathways determined by dynamic tractography. This innovative method animates the rapid neural propagations from a given cortical area via the underlying white matter tracts on diffusion-weighted image (DWI) tractography by quantifying the early neural responses to single-pulse electrical stimulation (SPES). We will compute the likelihood of postoperative seizure control using phase-amplitude coupling (PAC) between spontaneous high-frequency oscillation (HFO) at 80-300 Hz and slow-wave. Our innovative iEEG analysis will distinguish randomly-occurring pathological HFOs from physiological high gamma augmentation strictly time-locked to stimuli and behaviors. Our model's significance lies in its independence of the conventional electrical stimulation mapping, which may elicit nonhabitual seizures and fail to localize language areas satisfactorily in subsets of patients.

Sponsor Name: National Institute of Neurological Disorders and Stroke - NINDS (2R01NS064033)