Prediction of Stimulation-Defined Eloquent Cortex Using Graph-Theoretical Connectivity from Electrocorticography...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Imaging

Background:

• Electrical stimulation mapping (ESM) is the gold standard for identifying eloquent cortex but has limitations.
• Limitations include being time-intensive, incomplete cortical sampling, and patient tolerance issues.

Purpose of the Study:

• Investigate electrocorticography (ECoG) features, including functional connectivity, for identifying functionally critical cortex.
• Examine how predictive performance varies across functional domains, tasks, and data quantity.

Main Methods:

• Recorded ECoG from 14 epilepsy patients during speech production tasks.
• Used graph-theoretic functional connectivity features from high-gamma activity (70-150 Hz) and anatomical encoding.
• Trained machine learning classifiers and validated using leave-one-subject-out cross-validation with ESM-defined deficits as ground truth.

Main Results:

• Motor cortex electrodes were identified with the highest accuracy (ROC-AUC 0.929).
• Speech arrest prediction improved with more trials (plateauing around 10 trials).
• Language cortex prediction was less robust (ROC-AUC 0.761), with connectivity analysis offering the clearest improvement for speech arrest.

Conclusions:

• Graph-theoretic connectivity analysis of ECoG provides complementary information for presurgical mapping.
• Connectivity-informed approaches can help guide task selection and improve mapping efficiency.
• Combined-label prediction strategies offer a trade-off between broad detection and precise localization of critical cortex.