Epilepsy Prediction and Detection Using Attention-CssCDBN with Dual-Task Learning.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy affects millions globally, characterized by recurrent epileptic seizures.
• Electroencephalogram (EEG) monitoring is crucial for diagnosis but predicting seizures remains challenging.
• Variations within and between EEG signal classes complicate AI-driven epilepsy analysis.

Purpose of the Study:

• To develop an advanced AI model for accurate epilepsy prediction and detection.
• To address the challenge of intra-class and inter-class variations in EEG signals.
• To improve patient outcomes through timely seizure prediction and intervention.

Main Methods:

• Proposed a spatio-temporal EEGNet integrating a contractive slab and spike convolutional deep belief network (CssCDBN) with self-attention.
• Employed dual-task learning to extract high-order representations from EEG spectrum images, capturing spatial and temporal information.
• Utilized EEG-based verification during fine-tuning to reduce intra-class variation and enhance training efficiency.

Main Results:

• Achieved 98.5% sensitivity and a 0.041 false-positive rate (FPR) in epilepsy prediction.
• Demonstrated a prediction time of 50.92 minutes for impending seizures.
• Reached 94.1% accuracy in epilepsy detection, outperforming existing state-of-the-art methods.

Conclusions:

• The spatio-temporal EEGNet effectively extracts deep representations from EEG data for improved epilepsy analysis.
• The model significantly enhances the accuracy and timeliness of epileptic seizure prediction and detection.
• This AI-driven approach offers a promising advancement for managing epilepsy and improving patient care.