Epileptic seizure classifications using empirical mode decomposition and its derivative.

Area of Science:

• Neurology
• Signal Processing
• Machine Learning

Background:

• Epilepsy is a common neurological disorder characterized by disrupted brain activity.
• Electroencephalography (EEG) is crucial for classifying and detecting epileptic seizures.
• Empirical Mode Decomposition (EMD) and Ensemble EMD (EEMD) decompose complex EEG signals into intrinsic mode functions (IMFs).

Purpose of the Study:

• To develop and evaluate a hybrid IMF selection method for EEG signal analysis.
• To investigate the impact of selected IMFs from EMD and EEMD on epilepsy classification.
• To enhance the accuracy of detecting pre-seizure and seizure segments in EEG data.

Main Methods:

• A hybrid IMF selection method combining energy, correlation, power spectral distance, and statistical significance measures was developed.
• Multichannel EEG signals from epilepsy patients were decomposed into IMFs using EMD and EEMD.
• Time-domain, spectral-domain, and nonlinear features were extracted from selected IMFs for classification.

Main Results:

• The hybrid IMF selection approach significantly impacted classification results for both EMD and EEMD.
• Maximum classification accuracies reached up to 96.8% with EMD and 97% with EEMD across various classifiers.
• Classification performance using selected IMFs outperformed using direct EEG signals.

Conclusions:

• The proposed hybrid IMF selection method effectively enhances EEG-based epilepsy classification.
• EEMD, combined with the proposed selection method, offers a robust approach for feature extraction in seizure detection.
• The findings suggest improved accuracy in classifying pre-seizure and seizure states using advanced signal decomposition techniques.