Epileptic Seizure Detection with EEG Textural Features and Imbalanced Classification Based on EasyEnsemble Learning.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Automatic seizure detection from electroencephalogram (EEG) signals is crucial for epilepsy management.
• Classifying imbalanced data, where non-seizure periods vastly outweigh seizure events, presents a significant challenge in EEG analysis.

Purpose of the Study:

• To develop and validate an imbalanced learning model for improved seizure event identification in long-term EEG recordings.
• To enhance the accuracy and reliability of automatic seizure detection systems.

Main Methods:

• Feature extraction using discrete wavelet transform (DWT) and uniform 1D-Local Binary Patterns (LBP) to capture EEG signal characteristics.
• An ensemble learning framework combining weakly trained Support Vector Machines (SVMs) using under-sampling techniques.
• Multi-level decision fusion incorporating temporal and frequency factors for final seizure detection.

Main Results:

• Achieved high performance on public EEG databases, including an epoch-level F-score of 97.14% and event-level sensitivity of 96.67% on an intracranial database.
• Demonstrated a low false detection rate of 0.86/h and 0.81/h on long-term intracranial and scalp databases, respectively.
• Outperformed 14 published methods, highlighting improved detection performance and generalizability for imbalanced EEG data.

Conclusions:

• The proposed imbalanced learning model effectively addresses the challenge of classifying imbalanced EEG data for automatic seizure detection.
• The combination of DWT, LBP, ensemble SVMs, and multi-level fusion provides a robust and generalizable approach for seizure detection.
• The model shows significant potential for clinical application in epilepsy monitoring and management.