Supervised and Unsupervised Deep Learning Approaches for EEG Seizure Prediction.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy impacts over 50 million globally, characterized by unpredictable seizures.
• Seizure prediction could significantly reduce risks and stress for epilepsy patients.
• Current methods for seizure detection often lack precision, highlighting the need for advanced predictive tools.

Purpose of the Study:

• To develop and evaluate deep learning models for detecting pre-seizure (preictal) electroencephalogram (EEG) patterns.
• To compare the efficacy of supervised and unsupervised deep learning approaches in identifying preictal EEG.
• To explore the potential of anomalous event detection for seizure prediction using only normal EEG data.

Main Methods:

• Developed supervised deep learning models to classify preictal EEG from normal EEG.
• Developed novel unsupervised deep learning models to detect preictal EEG as an anomaly within normal EEG.
• Trained and evaluated models on two large, person-specific EEG seizure datasets.

Main Results:

• Both supervised and unsupervised deep learning approaches demonstrated feasibility in predicting seizures.
• Model performance varied based on individual patient characteristics, chosen approach, and specific deep learning architecture.
• The study confirmed the potential of deep learning for personalized seizure prediction.

Conclusions:

• Deep learning offers a viable pathway for developing seizure prediction systems.
• Personalized approaches are crucial, as model performance is patient-dependent.
• This research opens avenues for novel therapeutic interventions and improved epilepsy management, potentially saving lives.