An Automated Multivariate EEG Signal Analysis Framework for Seizure Detection Using D-MVMD and BO-SVM

Area of Science:

• Neurology
• Biomedical Engineering
• Signal Processing

Background:

• Epileptic seizures (ES) are a significant neurological disorder requiring accurate detection from electroencephalogram (EEG) signals for timely diagnosis and treatment.
• Automated ES detection is challenging due to the complex, noisy, and variable nature of EEG data.
• Existing methods struggle with the non-stationary, non-linear, and high-dimensional characteristics of EEG signals.

Purpose of the Study:

• To propose an automated framework for detecting epileptic seizures (ES) from multichannel electroencephalogram (EEG) signals.
• To address the challenges posed by noise, non-stationarity, and inter-subject variability in EEG data.
• To enhance the accuracy and robustness of automated seizure detection systems.

Main Methods:

• De-mixing Multivariate Variational Mode Decomposition (D-MVMD) was employed to decompose EEG signals into intrinsic mode functions, mitigating mode correlation.
• Multi-domain features capturing temporal, spectral, and non-linear dynamics were extracted from the decomposed signals.
• A Bayesian Optimized Support Vector Machine (BO-SVM) classifier, optimized using ReliefF-ranked random forest feature selection, was utilized for classification.

Main Results:

• The proposed framework achieved high performance metrics on the CHB-MIT dataset: 98.52% accuracy, 98.67% precision, 98.54% sensitivity, 98.54% specificity, and 0.98 F1 score.
• Robustness was confirmed through evaluation on the Siena dataset, assessing performance across different recording sessions of the same patient.
• Comparative analysis demonstrated superior mode separation and enhanced seizure detection capabilities compared to state-of-the-art methods.

Conclusions:

• The developed D-MVMD and BO-SVM integrated framework is an effective and robust solution for automated epileptic seizure detection using multichannel EEG analysis.
• The method successfully addresses the inherent complexities and noise in EEG signals, offering improved diagnostic potential.
• This approach provides a significant advancement in the field of automated neurological disorder detection.