Automatic diagnostics of electroencephalography pathology based on multi-domain feature fusion

Area of Science:

• Neuroscience and Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Electroencephalography (EEG) is a vital, non-invasive tool for diagnosing brain disorders due to its high temporal resolution and cost-effectiveness.
• Machine learning has advanced automated EEG diagnostics, but existing methods often overlook crucial spatial correlations and complementary features.
• Ignoring these factors and including redundant features can lead to overfitting and increased model complexity.

Purpose of the Study:

• To propose a novel feature-based framework to enhance the diagnostic accuracy of multi-channel EEG pathology detection.
• To address limitations in current automated EEG analysis by incorporating spatial information and reducing feature redundancy.

Main Methods:

• A multi-resolution decomposition and statistical feature extraction were used to create a time-frequency feature space.
• Spatial distribution information was extracted and fused with time-frequency features to leverage complementary data.
• A two-step dimension reduction strategy, including feature aggregation and statistical significance analysis, was employed to select discriminative features.

Main Results:

• The proposed framework demonstrated superior performance compared to state-of-the-art methods on the Temple University Hospital Abnormal EEG Corpus.
• The integration of spatial and temporal features, coupled with effective dimension reduction, significantly improved diagnostic accuracy.
• The method successfully identified and utilized features with stronger discriminative ability.

Conclusions:

• The novel feature-based framework effectively improves multi-channel EEG pathology diagnostic accuracy by considering spatial correlations and reducing feature redundancy.
• This approach holds significant potential for advancing clinically automated EEG abnormality detection.
• The findings underscore the importance of comprehensive feature engineering in machine learning for neurological diagnostics.