Abstracts

Functional Mapping of Language with High Gamma Electrocorticography

Abstract number : V.026
Submission category : 3. Neurophysiology / 3E. Brain Stimulation
Year : 2021
Submission ID : 1825732
Source : www.aesnet.org
Presentation date : 12/9/2021 12:00:00 PM
Published date : Nov 22, 2021, 06:50 AM

Authors :
Jennifer Shum, MD - Weill Cornell Medical Center; Patricia Dugan, MD - NYU Grossman School of Medicine; Daniel Friedman, MD - NYU Grossman School of Medicine; Werner Doyle, MD - NYU Grossman School of Medicine; Orrin Devinsky, MD - NYU Grossman School of Medicine; Adeen Flinker, PhD - NYU Grossman School of Medicine

Rationale: Electrical stimulation mapping (ESM) is the current gold standard for identifying eloquent language cortex which should be spared during epilepsy surgery. However, there are many limitations to this technique, particularly the risk of after discharges and seizures. ESM can also be time-consuming, requires excellent patient cooperation, and is not always well-tolerated by patients. High gamma electrocorticography (hgECoG) has been studied as another modality for pre-surgical language mapping as compared to ESM. HgECoG activity has been previously established as a robust marker of local cortical activity, making it an ideal candidate for functional mapping. However, existing studies comparing hgECoG to ESM have mixed results with highly variable sensitivities and specificities. It remains unclear what combination of hgECoG signal processing parameters and language tasks, as well as their spatial relationship, are most predictive of ESM results.

Methods: To overcome these limitations we utilize a battery of five language tasks and a statistical learning approach. Our language tasks capture multiple modalities of language processing and production and mirror the clinical paradigms employed during ESM. The tasks involved visual naming, visual word reading, auditory word repetition, auditory naming, and auditory sentence completion. Our statistical learning approach tests multiple supervised machine learning algorithms to predict which brain regions will be identified by ESM, and is based on hgECoG features during the five language tasks and normalized electrode spatial information.

Results: Using data from 12 subjects, the logistic regression algorithm performed well with a 5 fold cross validated AUC of 0.78 and had a specificity of 0.91 and sensitivity of 0.4 at its optimal operating point.

Conclusions: Our model shows that high gamma ECoG is a clinically useful tool to complement stimulation based language mapping.

Funding: Please list any funding that was received in support of this abstract.: This work is funded by Finding a Cure for Epilepsy and Seizures (FACES) and the American Epilepsy Society (AES) Research & Training Fellowship for Clinicians.

Neurophysiology