Seizure prediction based on improved vision transformer model for EEG channel optimization.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Epileptic seizures are unpredictable neurological events.
• Continuous electroencephalogram (EEG) monitoring is crucial for seizure prediction.
• Existing methods often require extensive data and computational resources.

Purpose of the Study:

• To develop a patient-specific seizure prediction method for lightweight, wearable devices.
• To extract robust features from multichannel EEG signals.
• To reduce the number of EEG channels required for accurate seizure prediction.

Main Methods:

• A vision transformer (ViT) based algorithm, Sel-JPM-ViT, was designed for seizure prediction.
• Time-frequency analysis was used to generate EEG spectrograms from multichannel EEG signals.
• A channel selection layer within the ViT model identified optimal channels for prediction.

Main Results:

• The Sel-JPM-ViT model achieved high performance using only 3-6 EEG channels, outperforming methods using all 22 channels.
• Average classification accuracy reached 93.65%, with 94.70% sensitivity and 92.78% specificity.
• The model demonstrated effectiveness on the Boston Children's Hospital-MIT scalp EEG dataset.

Conclusions:

• The Sel-JPM-ViT model offers an efficient and effective approach for patient-specific seizure prediction.
• Reduced channel usage makes the method suitable for developing practical, wearable seizure-prediction devices.
• This approach advances the potential for real-time seizure forecasting and management.