Automatic detection of abnormal EEG signals using multiscale features with ensemble learning.

Area of Science:

• Neurology
• Machine Learning
• Signal Processing

Background:

• Electroencephalogram (EEG) is a cost-effective tool for diagnosing neurological disorders.
• Machine learning shows potential for EEG abnormality detection but often overlooks feature redundancy and patient age.
• Existing methods struggle with feature redundancy and fail to leverage patient age for enhanced EEG analysis.

Purpose of the Study:

• To develop a novel framework for classifying EEG recordings as normal or abnormal.
• To address feature redundancy in EEG analysis by extracting hierarchical salient features.
• To improve EEG detection accuracy by integrating patient age information.

Main Methods:

• Extracted hierarchical salient features using a time-wise multi-scale aggregation strategy on optimal discrete wavelet transform coefficients.
• Fused multi-scale features with patient age information to enhance discriminative power.
• Employed ensemble learning classifiers including CatBoost, LightGBM, and random forest for classification.

Main Results:

• The proposed framework demonstrated superior performance compared to existing techniques when utilizing the CatBoost classifier.
• Integration of age information alongside multi-scale features improved the discrimination of EEG recordings.
• The method effectively identified significant EEG-derived features for pathology detection.

Conclusions:

• The developed framework offers a promising approach for EEG pathology detection by addressing feature redundancy and incorporating patient age.
• The findings highlight the importance of hierarchical feature extraction and age-informed analysis in clinical EEG.
• The study indicates significant potential for machine learning-based EEG analysis in neurological disease diagnosis.