Automated epilepsy diagnosis using interictal scalp EEG.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy affects over 50 million globally, with diagnosis often requiring lengthy electroencephalogram (EEG) recordings and expert analysis of seizure (ictal) activity.
• Collecting ictal EEG data is challenging, costly, and inconvenient, particularly in resource-limited regions.
• Existing automated systems often depend on ictal EEG, limiting their practical application.

Purpose of the Study:

• To develop and validate an automated system for epilepsy diagnosis using easily obtainable interictal scalp EEG data.
• To demonstrate the feasibility of diagnosing epilepsy from non-seizure EEG recordings.
• To improve accessibility of epilepsy diagnosis, especially in underserved areas.

Main Methods:

• Extraction of three distinct feature classes from interictal EEG data.
• Development of Probabilistic Neural Networks (PNNs) utilizing these extracted features.
• Optimization of feature extraction parameters and implementation of a voting mechanism to combine multiple PNNs.

Main Results:

• The automated system achieved a high diagnostic accuracy of 94.07%.
• The system successfully utilizes interictal EEG, which is simpler to collect than ictal EEG.
• The optimized feature extraction and PNN combination proved effective for epilepsy detection.

Conclusions:

• Interictal scalp EEG data can be effectively used for automated epilepsy diagnosis.
• The developed system offers a cost-effective and convenient alternative to traditional methods.
• This approach has the potential to significantly improve epilepsy diagnosis accessibility worldwide.