AI-Driven SEEG Channel Ranking for Epileptogenic Zone Localization.

Area of Science:

• Neuroscience
• Medical Technology
• Computational Biology

Background:

• Stereo-electroencephalography (SEEG) is crucial for pre-surgical epilepsy evaluation.
• Manual analysis of SEEG data from numerous channels is inefficient and time-consuming.

Purpose of the Study:

• To develop and validate a machine learning approach for ranking impactful SEEG channels.
• To enhance the efficiency and accuracy of pre-surgical epilepsy evaluation.

Main Methods:

• A classification model using XGBoost was trained to identify discriminative channel features during ictal periods.
• SHapley Additive exPlanations (SHAP) scoring was used to rank SEEG channels by seizure contribution.
• A channel extension strategy was implemented to identify potential epileptogenic zones beyond clinician selections.

Main Results:

• The machine learning approach demonstrated promising accuracy and consistency in ranking SEEG channels.
• SHAP analysis provided explainability for channel rankings, aiding clinical interpretation.
• The channel extension strategy successfully identified additional suspicious areas.

Conclusions:

• The proposed machine learning method offers an efficient and explainable tool for SEEG channel analysis in epilepsy surgery.
• This approach can improve the identification of epileptogenic zones, optimizing pre-surgical planning.
• Further validation across diverse patient cohorts is warranted.