Differentiating Epileptic and Psychogenic Non-Epileptic Seizures Using Machine Learning Analysis of EEG Plot Images.

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Imaging

Background:

• Epilepsy is a common neurological disorder where treatment failure is frequent.
• Non-epileptic seizures often complicate epilepsy treatment, necessitating accurate differentiation.
• Current methods for distinguishing seizure types, like electroencephalogram (EEG) analysis, are time-consuming and costly.

Purpose of the Study:

• To develop and validate a machine learning model for distinguishing epileptic from non-epileptic seizures using EEG images.
• To assess the accuracy and robustness of a convolutional neural network (CNN) approach in a real-world clinical setting.

Main Methods:

• Utilized a CNN with transfer learning (MobileNetV2) to analyze EEG plot images.
• Trained and cross-validated the model on 5359 EEG images from 107 adult subjects across two epilepsy monitoring units.
• Employed a visual analysis approach mimicking epileptologists' interpretation of EEG data.

Main Results:

• The CNN model achieved 86.9% accuracy (AUC 0.92) on training data and 87.3% accuracy (AUC 0.94) on validation data.
• Demonstrated high accuracy in distinguishing epileptic from non-epileptic seizures.
• Showcased the model's robustness across different EEG visualization software and medical facilities.

Conclusions:

• CNN analysis of EEG images is a highly accurate method for seizure classification.
• This approach offers a promising, potentially more accessible alternative to traditional EEG analysis for epilepsy diagnosis.
• The findings lay the foundation for advanced seizure subclassification using AI in clinical practice.