An EEG-based patient-independent epileptic seizure detection method based on domain generative adversarial network

Epilepsy is a prevalent chronic neurological disorder, and electroencephalogram (EEG) is a crucial tool for its diagnosis. However, the visual inspection of long-term EEG recordings is time-consuming and labor-intensive. Clinically, there is a need to detect epileptic seizures in patients not previously encountered, yet the EEG characteristics of seizures exhibit significant inter-patient variability.

This paper proposes a patient-independent epileptic seizure detection model based on a Domain Generative Adversarial Network (DGAN), which integrates two adversarial structures: a generative adversarial network and an adversarial domain adaptation network. This approach aims to reduce both inter-patient and intra-patient feature representation discrepancies, thereby achieving superior performance in patient-independent seizure detection.

The proposed method was evaluated on the publicly available CHB-MIT dataset and a proprietary dataset, demonstrating superior performance compared to existing methods. The segment-based evaluation achieved AUC scores of 0.8703 and 0.9107 on the two datasets, respectively, while the event-based evaluation achieved recall of 0.9392 and 0.9867, with false alarm rates of 3.47 and 2.47 per hour.

The results indicate that the proposed method is effective for patient-independent epileptic seizure detection. Moreover, this approach does not rely on patient identity labels, offering strong adaptability and scalability.