Patient-specific seizure detection method using nonlinear mode decomposition for long-term EEG signals.

Area of Science:

• Medical Technology
• Biomedical Signal Processing
• Neurology

Background:

• Automated seizure detection is crucial for modern medical development.
• Long-term electroencephalogram (EEG) recordings require efficient patient-specific analysis.
• Existing methods may lack robustness across diverse patient populations and seizure types.

Purpose of the Study:

• To explore a feasible approach for patient-specific seizure detection in long-term EEG.
• To introduce and evaluate a novel method based on nonlinear mode decomposition (NMD).
• To assess the efficacy of fractional central moments (FCM) for classifying seizure and non-seizure EEG signals.

Main Methods:

• A sliding window segmented multi-channel EEG into 2-s epochs.
• Nonlinear mode decomposition (NMD) decomposed EEG into nonlinear modes (NMs).
• Fractional central moments (FCM) were calculated on the first two NMs, forming feature vectors for KNN classification.

Main Results:

• The proposed FCM features with K nearest neighbor (KNN) demonstrated consistent high sensitivity across all patients.
• The method achieved average sensitivity of 98.40%, specificity of 99.10%, and accuracy of 98.61% on the CHB-MIT database.
• The approach proved reliable and robust to variations in seizure types among patients.

Conclusions:

• The FCM in the NMD domain is effective for classifying seizure and non-seizure EEG signals.
• The proposed method offers a reliable and robust single-feature diagnostic tool for seizures.
• This technique shows promising prospects for real-time automated seizure detection expert software applications.