Automatic Seizure Detection and Prediction Based on Brain Connectivity Features and a CNNs Meet Transformers...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy is a neurological disorder characterized by recurrent seizures.
• Electroencephalogram (EEG) patterns vary across seizure states (inter-ictal, pre-ictal, ictal).
• Brain connectivity networks, a 2D feature, are underutilized for seizure detection and prediction.

Purpose of the Study:

• To investigate the effectiveness of brain connectivity features for epilepsy seizure detection and prediction.
• To develop subject-specific (SSM), subject-independent (SIM), and cross-subject (CSM) models for seizure analysis.
• To evaluate the performance and efficiency of different connectivity measures and frequency bands.

Main Methods:

• Extracted image-like brain connectivity features using various time-window lengths, frequency bands (e.g., β, γ), and connectivity measures (e.g., Pearson Correlation Coefficient, Phase Lock Value).
• Employed a support vector machine (SVM) for SSM and a Convolutional Neural Networks meet Transformers (CMT) classifier for SIM and CSM.
• Conducted feature selection and efficiency analyses on the CHB-MIT dataset.

Main Results:

• Longer time windows generally improved performance.
• Achieved high detection accuracies: 100.00% (SSM), 99.98% (SIM), and 99.27% (CSM).
• Reached high prediction accuracies: 99.72% (SSM), 99.38% (SIM), and 86.17% (CSM).
• Pearson Correlation Coefficient and Phase Lock Value in β and γ bands demonstrated strong performance and efficiency.

Conclusions:

• The proposed brain connectivity features are reliable and valuable for automated seizure detection and prediction.
• These findings support the development of portable, real-time epilepsy monitoring equipment.
• Brain connectivity analysis offers a promising avenue for advancing epilepsy management.