Automated Detection of High-Frequency Oscillations in Epilepsy Based on a Convolutional Neural Network.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy is a common chronic neurological disorder.
• High-frequency oscillations (HFOs) are key biomarkers for the epileptogenic zone.
• Manual HFO detection is time-consuming and subjective.

Purpose of the Study:

• To develop and validate a convolutional neural network (CNN) for automated detection of HFOs (ripples and fast ripples).
• To compare the CNN's performance against existing HFO detection methods.
• To assess the potential clinical utility of the CNN for identifying the epileptogenic zone.

Main Methods:

• Intracranial electroencephalograms (iEEG) from six epilepsy patients were analyzed.
• A convolutional neural network (CNN) was employed for detecting ripples and fast ripples.
• Performance was evaluated against four other HFO detectors in RIPPLELAB and visual analysis.

Main Results:

• The CNN detector achieved higher sensitivity (77.04% ripples, 83.23% fast ripples) and specificity (72.27% ripples, 79.36% fast ripples).
• Cohen's kappa coefficients indicated good agreement with visual analysis (0.541 ripples, 0.777 fast ripples).
• The CNN outperformed four other automated HFO detection methods.

Conclusions:

• The developed CNN method provides accurate and reliable automated detection of HFOs.
• This automated detector demonstrates superior performance compared to existing methods.
• The CNN holds promise for assisting clinicians in precise epileptogenic zone localization.