A multi-view neural framework with attention for epileptic seizure classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy is a neurological disorder marked by recurrent seizures.
• Electroencephalogram (EEG) analysis is crucial for seizure diagnosis.
• Automated EEG-based seizure classification aids clinical decision-making.

Purpose of the Study:

• To develop an automated system for epileptic seizure classification using EEG signals.
• To reduce reliance on subjective expert interpretation in epilepsy diagnosis.
• To enhance the accuracy of automated seizure detection and classification.

Main Methods:

• Proposed a novel filter-bank multi-view and attention-based neural network model (FB-AMNet).
• Employed a learnable filter bank for multi-view EEG decomposition into frequency sub-bands.
• Utilized multi-branch group convolution and bidirectional LSTM with attention for feature extraction.

Main Results:

• Achieved an overall F1 score of 0.7105 and weighted F1 score of 0.8314 on the TUSZ dataset.
• Demonstrated significant improvements over the baseline FBCNet method (p < 0.05).
• Obtained optimal results on both TUSZ and CHB-MIT datasets.

Conclusions:

• The FB-AMNet model effectively classifies epileptic seizures using EEG data.
• Combining multi-view feature extraction with attention-enhanced temporal modeling improves classification performance.
• The proposed method shows promise for clinical application in epilepsy diagnosis.