EEG-TGC: A Novel Self-Supervised Method based on Temporal-Graph Contrast for Seizure Detection.

Area of Science:

• Neurology
• Machine Learning
• Biomedical Signal Processing

Background:

• Epilepsy diagnosis relies on electroencephalography (EEG), but manual annotation is time-consuming.
• Existing seizure detection methods often underutilize unlabeled EEG data and spatial electrode information.
• Challenges include the significant burden of data annotation and suboptimal classification due to ignored spatial distributions.

Purpose of the Study:

• To develop a novel self-supervised learning method for seizure detection that effectively utilizes unlabeled EEG data.
• To incorporate spatial topological information from multi-electrode EEG recordings.
• To reduce the annotation burden on clinicians while maintaining high seizure detection performance.

Main Methods:

• Proposed EEG-TGC, a self-supervised learning method based on temporal graph contrast.
• Introduced node and graph contrast to capture spatial topological information from EEG graphs.
• Evaluated the method on the large public TUSZ EEG dataset.

Main Results:

• EEG-TGC achieved performance comparable to supervised learning using 100% labeled data.
• The method demonstrated high efficacy even with only 10% labeled data.
• Successfully utilized unlabeled EEG data and captured spatial electrode distribution information.

Conclusions:

• The developed algorithm offers an effective solution for automatic seizure detection.
• EEG-TGC significantly reduces the need for extensive manual annotation of EEG recordings.
• The method provides a promising approach for epilepsy diagnosis with limited labeled data.