M2D2: Maximum-Mean-Discrepancy Decoder for Temporal Localization of Epileptic Brain Activities.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Deep learning models for epilepsy monitoring using electroencephalographic (EEG) signals show limited generalization across different clinical settings.
• Manual labeling of EEG data is time-consuming and requires expert analysis, hindering patient-specific model adaptation.

Purpose of the Study:

• To develop an automated method for temporal localization and labeling of seizures in long EEG recordings.
• To improve the generalization performance of deep learning models for epilepsy monitoring.

Main Methods:

• Proposed the Maximum-Mean-Discrepancy Decoder (M2D2) for automatic seizure detection in EEG signals.
• Evaluated M2D2's performance on EEG data from a different clinical setting than the training data.

Main Results:

• M2D2 achieved an F1-score of 76.0% for temporal localization on out-of-setting data.
• M2D2 achieved an F1-score of 70.4% for temporal localization on out-of-setting data.
• Demonstrated substantially higher generalization performance compared to state-of-the-art deep learning approaches.

Conclusions:

• M2D2 offers a promising solution for automatic seizure detection and labeling in EEG recordings.
• The proposed method enhances model generalization, reducing the need for extensive manual labeling and expert re-analysis.
• M2D2 can assist medical experts in epilepsy monitoring, particularly in diverse clinical environments.