Machine learning (ML) in intraoperative neuromonitoring (IONM): proof of concept

Area of Science:

• Neurosurgery
• Machine Learning
• Spinal Deformity Correction

Background:

• Intraoperative neuromonitoring (IONM) is crucial for pediatric spine surgery safety.
• Current IONM interpretation relies on human expertise and can be variable.
• Subtle neurophysiological changes may precede neurological injury, necessitating advanced detection methods.

Purpose of the Study:

• To evaluate a machine learning (ML) algorithm for early detection of neural compromise during pediatric spine surgery.
• To assess the ML algorithm's ability to identify subtle changes in motor evoked potentials (MEPs).
• To compare ML-based alerts with standard clinical intraoperative neuromonitoring alerts.

Main Methods:

• Retrospective analysis of IONM data from 84 pediatric spine surgeries.
• Training an ML model on baseline muscle-specific MEPs to detect intraoperative signal changes.
• Continuous analysis of real-time MEP data with a "red flag" alert for deviations exceeding 10% per minute.

Main Results:

• The ML model achieved 78.6% sensitivity and 69.0% overall accuracy (AUC=0.78).
• ML-generated alerts preceded clinical IONM alerts by an average of 23.3 minutes.
• The model correctly flagged 5 out of 6 patients who developed postoperative deficits.

Conclusions:

• The ML-based tool shows potential for early detection of IONM changes during pediatric spine surgery.
• High sensitivity and negative predictive value suggest utility for real-time surgical support.
• Further validation in larger, prospective studies is warranted.