Detecting Cerebral Ischemia From Electroencephalography During Carotid Endarterectomy Using Machine Learning.

Area of Science:

• Neuroscience
• Medical Technology
• Artificial Intelligence

Background:

• Cerebral ischemia detection during surgery using electroencephalography (EEG) is crucial for preventing stroke.
• Current neurophysiologist-based monitoring is susceptible to errors, necessitating improved diagnostic methods.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models for accurate and efficient intraoperative ischemia detection.
• To compare the performance of ML models against human neurophysiologists in identifying cerebral ischemia.

Main Methods:

• Supervised ML models were trained on a dataset of 802 patients with intraoperative ischemia.
• Models were validated on a separate dataset of 30 patients, with labels refined by five neurophysiologists.
• Performance metrics included sensitivity, specificity, Cohen's kappa, AUROC, and AUPRC.

Main Results:

• Neurophysiologist agreement showed moderate-to-substantial agreement (kappa: 0.59–0.74).
• ML models achieved comparable sensitivity (63–89%) and specificity (85–96%) to neurophysiologists (58–93% and 83–96%, respectively).
• Random Forest, LightGBM, and XGBoost demonstrated high performance with AUROC values of 0.92–0.93 and AUPRC values of 0.79–0.83.

Conclusions:

• ML models offer a reliable and potentially more accurate approach to intraoperative ischemia monitoring.
• Implementing ML can enhance patient safety during surgery by improving stroke precursor detection.
• This technology has the potential to reduce healthcare costs associated with monitoring and stroke complications.