Expert-Level Intracranial Electroencephalogram Ictal Pattern Detection by a Deep Learning Neural Network.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Interpreting intracranial EEG (iEEG) is crucial for epilepsy surgery decisions.
• Deep learning shows promise for EEG analysis, but iEEG seizure detection is limited by small patient-specific datasets.
• Responsive neurostimulation (RNS) systems generate large iEEG datasets, ideal for developing advanced detection algorithms.

Purpose of the Study:

• To evaluate the efficacy of a deep learning methodology for detecting iEEG seizures.
• To assess the performance of a convolutional neural network (CNN) using a large dataset from RNS systems.
• To determine the potential for automated seizure detection in improving closed-loop brain stimulation management.

Main Methods:

• Collected 5,226 ictal events from 22 patients with RNS implants.
• Developed a personalized CNN for seizure annotation, tested in two scenarios: post-chronic recording and post-implantation.
• Evaluated CNN accuracy against human neurophysiologists using area-under-precision-recall curve (AUPRC) and regression accuracy.

Main Results:

• The CNN achieved high accuracy in both scenarios, with mean AUPRC of 0.84 (scenario 1) and 0.80 (scenario 2).
• Mean regression accuracy was 6.3 seconds in both scenarios, with near-maximum performance at 10 seed samples.
• Classification failures were linked to specific EEG patterns like electro-decrements, brief seizures, and sleep-wake cycle changes.

Conclusions:

• A deep learning neural network was developed for personalized detection of RNS-derived ictal patterns with expert-level accuracy.
• Automated seizure detection holds significant potential for enhancing closed-loop brain stimulation management.
• This approach is effective even during the initial recording period when the RNS system is learning patient-specific seizure characteristics.