An event-based filtering and weighted enhanced deep learning epileptic seizure prediction method.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Timely seizure detection is crucial for epilepsy management, enabling prompt medical interventions and patient coping strategies.
• Traditional electroencephalogram (EEG) signal analysis for seizure prediction often lacks temporal continuity, limiting clinical applicability.
• Event-based prediction is preferred over segment-based classification for improved practical usefulness in real-time seizure management.

Purpose of the Study:

• To develop and evaluate an innovative two-step approach for enhanced seizure prediction.
• To address the limitations of traditional segment-based EEG analysis by incorporating event-based detection.
• To improve the temporal continuity and practical utility of seizure prediction models.

Main Methods:

• A two-step seizure prediction framework was implemented, combining segment-based and event-based prediction.
• The first step utilized a Particle Swarm Optimization (PSO) optimized Dual Attention Mechanism (DAM) integrated with a 2D Convolutional Neural Network (2DCNN) for segment-based prediction.
• The second step employed a two-layer 'k-of-n' logic filter for effective seizure event detection.

Main Results:

• The proposed PSO-DAM-2DCNN model demonstrated strong performance on segment-based metrics.
• The integrated two-step approach showed promising results in event-based metrics, including False Positive Rate per hour (FPR/h), False Negative Rate (FNR), and True Positive Rate (TPR).
• The method was validated on both the CHB-MIT and Huashan Hospital private datasets, indicating generalizability.

Conclusions:

• The novel two-step seizure prediction method offers a significant advancement over traditional approaches.
• The integration of PSO, DAM, 2DCNN, and k-of-n logic filter provides a robust framework for accurate and temporally continuous seizure prediction.
• This approach holds potential for improving clinical management and patient outcomes in epilepsy care.