Multi-branch convolutional neural network and intracranial EEG high-frequency oscillations predict post-surgical...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Technology

Background:

• Pathological high-frequency oscillations (HFOs) are biomarkers for epileptogenic cortex.
• Current epilepsy surgery planning relies on multidisciplinary consensus, with surgical utility of HFOs unproven.

Purpose of the Study:

• To test if a Convolutional Neural Network (CNN) can predict seizure freedom by integrating HFO features, neuroanatomy, and surgical boundaries.
• To evaluate the CNN model's accuracy in predicting post-operative seizure outcomes.

Main Methods:

• A three-branch CNN was trained using SEEG data from 78 pre-surgical epilepsy patients.
• Inputs included stereotaxic coordinates, resection status, and 37 HFO features.
• The model predicted post-operative seizure freedom probability using spatial, electrophysiological, and surgical data.

Main Results:

• The HFO-informed CNN model achieved 92% accuracy in predicting seizure-free patients via fivefold cross-validation.
• Fast ripples, particularly those on epileptiform spikes, were identified as key HFO features.

Conclusions:

• A CNN model integrating HFOs, neuroanatomy, and surgical data accurately predicts seizure freedom post-surgery.
• This AI approach can guide surgical planning and explore virtual resections to optimize outcomes.