Improving the accuracy of epileptogenic zone localization in stereo EEG with machine learning algorithms

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Technology

Background:

• Precise identification of the epileptogenic zone (EZ) is critical for successful epilepsy surgery.
• Stereo EEG analysis for EZ localization presents significant challenges for even expert epileptologists.
• Development of machine learning (ML)-based tools is essential for improving EZ localization accuracy.

Purpose of the Study:

• To develop and evaluate ML models for accurate EZ localization using Stereo EEG data.
• To compare the performance of different ML algorithms and feature types for EZ identification.

Main Methods:

• Developed ML models using Stereo EEG data from 15 epilepsy patients with successful surgical outcomes.
• Identified spikes and High Frequency Oscillations (HFOs) in the resected and non-resected zones.
• Extracted linear and non-linear features from spike and HFO data to train Gradient Boosting, Extra Trees, and Random Forest classifiers.

Main Results:

• Gradient Boosting achieved 98.5% accuracy for EZ localization in Mesial Temporal Lobe Epilepsy (MTLE) using spike-ripple features.
• Extra Trees achieved 87.6% accuracy for Neocortical Epilepsy (NE) using fast ripple features.
• Linear features demonstrated superior performance over non-linear features in predicting EZs; Random Forest, Extra Trees, and Gradient Boosting were most effective for RZ prediction.

Conclusions:

• ML methodologies can accurately localize epileptogenic zones, demonstrating their capability in presurgical epilepsy evaluation.
• Further improvements in accuracy are anticipated with larger datasets and advanced ML algorithms.
• Multi-center implementation is recommended for broader validation and generalizability of the ML approach.