EEG-Based Inter-Patient Epileptic Seizure Detection Combining Domain Adversarial Training with CNN-BiLSTM Network.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Automated epileptic seizure detection from electroencephalogram (EEG) is crucial but challenging due to high inter-patient variability.
• Existing patient-specific models struggle with generalization to new individuals.
• Developing robust, generalizable seizure detection methods is a significant unmet need.

Purpose of the Study:

• To develop and evaluate a novel deep learning framework for automated epileptic seizure detection.
• To improve the generalizability of seizure detection models across different patients.
• To combine domain adversarial training with CNN and BiLSTM for robust cross-patient seizure detection.

Main Methods:

• A convolutional neural network (CNN) was employed to extract local patient-invariant features via domain adversarial training.
• Domain adversarial training minimized patient-specific characteristics while optimizing seizure detection.
• A bidirectional long short-term memory (BiLSTM) network captured temporal dependencies for modeling seizure evolution.

Main Results:

• The proposed framework demonstrated superior performance compared to non-adversarial methods in cross-patient seizure detection.
• High detection accuracy was achieved across diverse patient cohorts with focal epilepsy.
• The integration of adversarial training and temporal modeling led to robust performance.

Conclusions:

• The developed framework effectively addresses the challenge of individual differences in EEG patterns for seizure detection.
• Domain adversarial training combined with BiLSTM offers a promising approach for generalizable automated epilepsy diagnosis.
• This method holds potential for improving clinical decision-making in epilepsy management.