Epileptic Seizure Classification with Patient-level and Video-level Contrastive Pretraining.

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate epileptic seizure classification requires specialized clinical expertise.
• Limited labeled clinical videos hinder the development of advanced seizure classification models.
• Existing action recognition modules face challenges due to data scarcity.

Purpose of the Study:

• To develop a transformer-based model for seizure classification using unlabeled clinical videos.
• To leverage contrastive learning for pretraining models without manual annotation.
• To differentiate between temporal lobe epilepsy (TLE) and extratemporal lobe epilepsy (exTLE).

Main Methods:

• Utilized unlabeled clinical videos for model pretraining.
• Employed a transformer-based architecture with contrastive loss.
• Maximized intra-patient/video similarity and minimized inter-patient/video similarity.
• Finetuned a classification head for TLE vs. exTLE discrimination.

Main Results:

• Achieved 5-fold cross-validation accuracy of 0.93 and an F1 score of 0.88 on video-level classification.
• Demonstrated superior performance compared to state-of-the-art seizure classification models.
• Successfully classified temporal lobe epilepsy (TLE) and extratemporal lobe epilepsy (exTLE).

Conclusions:

• Pretraining with unlabeled data using contrastive loss is effective for seizure classification.
• The proposed method reduces reliance on expert-annotated clinical data.
• This approach shows significant potential for improving clinical seizure diagnosis and patient care.