Epileptic Seizure Detection by Cascading Isolation Forest-Based Anomaly Screening and EasyEnsemble

Area of Science:

• * Neuroscience
• * Machine Learning
• * Medical Informatics

Background:

• * Electroencephalogram (EEG) is crucial for automatic epilepsy seizure detection.
• * Current algorithms require extensive labeled data, making manual labeling time-consuming.
• * A need exists for efficient seizure detection methods that minimize data annotation workload.

Purpose of the Study:

• * To develop a hybrid unsupervised learning (UL) and supervised learning (SL) system for epilepsy seizure detection.
• * To reduce the manual data labeling burden in training seizure detection algorithms.
• * To improve the efficiency and accuracy of automatic seizure detection.

Main Methods:

• * A hybrid system integrating UL and SL modules for seizure detection.
• * UL module: aEEG extraction, isolation forest, adaptive segmentation, and silhouette coefficient for preliminary screening.
• * SL module: EasyEnsemble algorithm for robust detection of potential seizure candidates identified by UL.

Main Results:

• * Achieved a mean accuracy of 92.62%, mean sensitivity of 95.55%, and mean specificity of 92.57% on the CHB-MIT dataset.
• * The hybrid approach significantly reduced the workload of data labeling.
• * Demonstrated competitive performance compared to state-of-the-art methods.

Conclusions:

• * The proposed hybrid UL-SL system effectively detects epilepsy seizures with high accuracy.
• * This method substantially decreases the need for manual data annotation in EEG analysis.
• * Represents a novel and effective approach for automated epilepsy seizure detection.