Towards an Online Seizure Advisory System-An Adaptive Seizure Prediction Framework Using Active Learning Heuristics.

Area of Science:

• * Neuroscience
• * Biomedical Engineering
• * Machine Learning

Background:

• * Seizure prediction systems offer significant potential for improving the quality of life for epilepsy patients.
• * Current prediction algorithm accuracy is limited by inherent brain signal uncertainty and recording artifacts.
• * Active learning is traditionally used to address ambiguity but selecting data for expert annotation remains challenging.

Purpose of the Study:

• * To develop an active learning-based seizure prediction framework.
• * To improve prediction accuracy using a minimal amount of labeled data.
• * To enhance medical reliability through efficient expert intervention.

Main Methods:

• * Employed a Bernoulli-Gaussian Mixture Model (BGMM) to identify the most ambiguous feature samples for expert annotation.
• * Evaluated seven different classifiers based on classification time and memory requirements.
• * Developed an active learning framework utilizing the best-performing classifier.

Main Results:

• * The proposed active learning approach achieved comparable accuracy to Support Vector Machine (SVM) classifiers using only 20% of labeled data.
• * Demonstrated improved prediction accuracy, even under noisy conditions.
• * Showcased reduced annotation effort required for high prediction accuracy.

Conclusions:

• * The BGMM-based active learning framework effectively reduces the need for extensive data labeling in seizure prediction.
• * This approach enhances the efficiency and reliability of seizure prediction systems.
• * The framework shows promise for real-world clinical applications requiring accurate and robust seizure detection.