Combined Anomaly Features and Interictal Epileptiform Discharges for Effective Seizure Onset Zone Localization

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate seizure onset zone (SOZ) identification is crucial for epilepsy surgery.
• Traditional SOZ localization relies heavily on neurosurgeon expertise.
• Automated methods using artificial intelligence are emerging for SOZ localization.

Purpose of the Study:

• To introduce and evaluate autoencoder-based anomaly detection for SOZ electrode classification.
• To investigate the efficacy of anomaly features, interictal epileptiform discharge (IED) biomarkers, and their combination for SOZ localization.

Main Methods:

• Trained an autoencoder using leave-one-patient-out cross-validation on electrocorticography (ECoG) signals.
• Calculated anomaly features based on reconstruction errors from the autoencoder.
• Classified SOZs using a linear support vector machine (SVM) with anomaly features, IED features, or combined features.

Main Results:

• Anomaly features alone achieved 70.49% accuracy in SOZ localization.
• IED features alone achieved 64.71% accuracy.
• Combining anomaly and IED features resulted in the highest accuracy of 74.44%.

Conclusions:

• Autoencoder-based anomaly detection is effective for direct SOZ localization.
• Combining multiple biomarker features significantly enhances automatic SOZ localization performance.
• This approach holds promise for improving surgical planning in epilepsy treatment.