Differential quadruple pattern: A new EEG signal classification framework.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalography (EEG) signals reflect neural activity, with abnormal patterns indicating brain disorders like epilepsy.
• Machine learning advancements have enabled automated and accurate EEG interpretation.
• Existing models often lack interpretability, hindering clinical adoption.

Purpose of the Study:

• To develop an explainable EEG classification model using novel feature engineering.
• To introduce and validate the Differential Quadruple Pattern (DiffQuadPat) feature extractor.
• To enhance the transparency of automated EEG analysis for brain disorder detection.

Main Methods:

• A novel feature extractor, DiffQuadPat, was developed, utilizing difference-based transformations and combinational transition tables.
• Feature selection was performed using Cumulative Weight Neighborhood Component Analysis (CWNCA).
• Classification was achieved with t-algorithm-based k-Nearest Neighbors (tkNN), and interpretability was enhanced using Directed Lobish (DLOB).

Main Results:

• The DiffQuadPat-centric XFE framework achieved over 98% accuracy on ALS and neonatal epilepsy detection datasets via 10-fold cross-validation.
• The model generated symbolic explanations and visualized cortical/hemispheric connectome diagrams.
• Transparent visualization of brain-level interactions was enabled.

Conclusions:

• The proposed explainable EEG classification model demonstrates high accuracy and interpretability.
• DiffQuadPat offers a promising approach for automated EEG analysis in clinical settings.
• The framework facilitates transparent understanding of brain activity related to neurological disorders.