Detecting fast-ripples on both micro- and macro-electrodes in epilepsy: A wavelet-based CNN detector

Area of Science:

• Neuroscience
• Medical Technology

Background:

• Fast-ripples (FR) are high-frequency oscillations (200-600 Hz) crucial for identifying the epileptogenic zone in epilepsy.
• Detecting FR in intracerebral EEG (iEEG) is challenging due to varying recording scales (macro- and microwires).

Purpose of the Study:

• To develop and validate a novel, efficient method for detecting fast-ripples (FR) in intracerebral EEG (iEEG).
• To ensure the method's effectiveness across both macro- and microwire iEEG recordings.

Main Methods:

• A convolutional neural network (CNN) was trained on over 11,000 FR events from 38 epilepsy patients.
• The detection pipeline includes CNN-based event identification, feature-based false positive control, and human validation via a GUI.
• The method, WALFRID, processes raw iEEG data without pre-processing steps like artifact rejection.

Main Results:

• WALFRID demonstrated high performance with up to 99.95% sensitivity and 96.51% precision on simulated data.
• The detector successfully adapted to both macro- and microwire iEEG recordings.
• Human validation effectively eliminated remaining false positives, requiring minimal time per subject.

Conclusions:

• WALFRID offers a user-friendly tool for neurologists, mimicking their analytical workflow.
• The method is easily usable, understandable, and correctable by clinicians.
• WALFRID performs comparably or superiorly to existing FR detection methods.