Single-Channel EEG-Based Epileptic Seizure Prediction Using Common Spatial Pattern and Transfer Learning.

Area of Science:

• Neurology
• Biomedical Engineering
• Signal Processing

Background:

• Epilepsy is a neurological disorder causing recurrent seizures, impacting patient safety and quality of life.
• Current electroencephalogram (EEG)-based seizure detection methods often require multiple channels, leading to discomfort and complexity.
• Developing efficient and patient-specific seizure prediction is crucial for improved patient management.

Purpose of the Study:

• To propose and evaluate a patient-specific, single-channel seizure prediction model for epilepsy.
• To reduce patient discomfort and computational load associated with traditional multichannel EEG methods.
• To demonstrate the feasibility of using advanced signal processing and deep learning for accessible seizure detection.

Main Methods:

• Utilized the CHB-MIT EEG dataset for model development and validation.
• Employed Common Spatial Patterns (CSP) for optimal single-channel selection.
• Applied Continuous Wavelet Transform (CWT) scalograms for feature extraction.
• Developed a ResNet50 deep learning model to differentiate between preictal (seizure imminent) and interictal (seizure-free) states.

Main Results:

• Achieved an average prediction accuracy of 85.1±3.2%.
• Reported an average sensitivity of 84.0±3.9% and specificity of 87.4±2.8% across 13 patient cases.
• Demonstrated the model's effectiveness on a patient-specific basis using single-channel EEG data.

Conclusions:

• The proposed single-channel seizure prediction model shows significant potential for developing wearable and accessible epilepsy monitoring systems.
• The method offers a less invasive and computationally efficient alternative to existing multichannel EEG approaches.
• Future research should focus on validating this approach on larger, diverse datasets and exploring patient-independent models for broader applicability.