Epileptic Seizure Detection and Prediction in EEGs Using Power Spectra Density Parameterization

Area of Science:

• Neuroscience
• Signal Processing
• Medical Diagnostics

Background:

• Electroencephalography (EEG) power spectrum analysis is vital for epilepsy classification.
• Conflation of periodic and aperiodic EEG signals can hinder epilepsy detection and prediction accuracy.

Purpose of the Study:

• To investigate the distinct roles of periodic and aperiodic EEG power spectrum components in epilepsy detection and prediction.
• To assess the efficacy of separating these components for improved diagnostic accuracy.

Main Methods:

• Utilized a power spectrum density parameterization method to differentiate periodic and aperiodic EEG signal components.
• Validated the approach on two public datasets: the Bonn EEG database and the CHB-MIT Long-term EEG database.

Main Results:

• Aperiodic components achieved higher classification accuracy (96.68%) than periodic components (73.9%) on the Bonn dataset.
• Combined features yielded 98.88% accuracy on the Bonn dataset and 99.95% detection accuracy on the CHB-MIT dataset.
• Successfully predicted all seizures in the CHB-MIT dataset with minimal false predictions.

Conclusions:

• Both periodic and aperiodic EEG components aid in epilepsy stage discrimination.
• Aperiodic neural activity plays a decisive role in epilepsy classification.
• Findings offer significant implications for enhancing epilepsy diagnosis accuracy and efficiency.