Artificial intelligence for intraoperative neuromonitoring: signal interpretation, risk prediction, and clinical translation

This reviews how artificial intelligence and machine learning reshape intraoperative neuromonitoring for thyroid and head and neck surgery with emphasis on protecting the recurrent laryngeal nerve. We synthesize four methodological strands including end-to-end deep learning on electromyography, classical machine learning with engineered features, motor evoked potential analytics, and computer vision for nerve localization. We map inputs, model classes, and objectives, and compare recurrent laryngeal nerve palsy prediction pipelines that use intraoperative electromyography trend dynamics, registry-based clinical ensembles, and voice spectrogram-derived outcomes. For real-time safety, we contrast threshold-based alerts with machine learning detectors and hybrid systems, and we highlight interpretability, acquisition to alert latency, and robustness to artifacts. Evidence includes prospective evaluations within operating room workflows, yet gaps remain in external validation and generalization across sites. We outline deployment principles that include calibrated graded alerts, standardized visualization, and surgeon-in-the-loop operation aligned with Standard Protocol Items: Recommendations for Interventional Trials–Artificial Intelligence (SPIRIT-AI), the Consolidated Standards of Reporting Trials–Artificial Intelligence (CONSORT-AI), and Developmental and Exploratory Clinical Investigations of DEcision support systems driven by Artificial Intelligence (DECIDE-AI). Together these elements enable earlier and more reliable detection and risk stratification while preserving clinical transparency.