Semisupervised Seizure Prediction in Scalp EEG Using Consistency Regularization.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy seizure prediction is crucial for patient well-being.
• Deep learning models for seizure prediction typically require extensive labeled electroencephalogram (EEG) data.
• Labeling EEG data is time-consuming and requires specialized expertise.

Purpose of the Study:

• To develop a deep learning model for epilepsy seizure prediction that minimizes the need for labeled data.
• To leverage unlabeled EEG data to improve seizure prediction accuracy.
• To reduce the cost and time associated with expert data labeling in seizure prediction.

Main Methods:

• Proposed a novel Consistency-based Semisupervised Seizure Prediction Model (CSSPM).
• Utilized consistency regularization principle: a robust model should yield consistent predictions under perturbations.
• Employed stochastic augmentation and dropout to create a stochastic neural network, enforcing consistency constraints between predictions.

Main Results:

• Achieved satisfactory seizure prediction performance with only a fraction of labeled training data.
• Demonstrated effective utilization of unlabeled EEG data to enhance the decision boundary.
• Outperformed traditional supervised methods that require fully labeled datasets.

Conclusions:

• The CSSPM offers a promising solution for epilepsy seizure prediction by significantly reducing labeling costs.
• Semisupervised learning, particularly consistency regularization, is effective for EEG-based seizure prediction.
• The model alleviates a major bottleneck in real-world clinical applications of seizure prediction technology.