An Attention-Based Hybrid Deep Learning Approach for Patient-Specific, Cross-Patient, and Patient-Independent Seizure...

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Automatic epilepsy detection is vital for patient diagnosis and treatment.
• Current patient-specific models face limitations in clinical application, especially with new patient data.
• Epileptic Seizure Detection (ESD) in electroencephalogram (EEG) data requires robust methods.

Purpose of the Study:

• To introduce a novel hybrid deep learning approach for EEG epileptic seizure detection.
• To develop a model capable of patient-specific, cross-patient, and patient-independent seizure detection.
• To enhance the accuracy and reliability of automatic epilepsy diagnosis.

Main Methods:

• A hybrid deep learning model combining a 1D Convolutional Neural Network (1D CNN) and a Multi-Long Short-Term Memory Network (MLSTM) with a Multi-Attention layer (MAT).
• The 1D CNN extracts spatial features from EEG data.
• The MLSTM extracts temporal features, and the MAT layer performs feature fusion and pattern identification.

Main Results:

• The proposed hybrid model demonstrated superiority over existing methods in EEG epileptic seizure detection.
• Achieved an average improvement of 2% in classification accuracy, recall, specificity, and G.mean.
• Validated effectiveness across patient-specific, cross-patient, and patient-independent detection scenarios.

Conclusions:

• The developed hybrid deep learning framework is robust and effective for EEG epileptic seizure detection.
• The novel approach offers significant improvements for clinical applications.
• This method provides a promising advancement in the field of automated neurological disorder diagnosis.