Artificial Intelligence in Operating Room Workflow Optimization

Abstract

Operating room scheduling and throughput have traditionally relied on historical averages and clinical heuristics that perform poorly at predicting the outliers that drive costly delays and cancellations. This review examines the emerging evidence for artificial intelligence applications across three domains of operating room workflow: machine learning models for surgical case-duration prediction, computer-vision and depth-sensing systems for real-time phase and milestone recognition, and anesthesia-specific predictive analytics for intraoperative hemodynamic and complication risk. Case-duration prediction is the most mature of these applications, with multicenter machine learning models consistently outperforming surgeon-estimate and moving-average baselines, while video- and depth-based workflow recognition is advancing rapidly but remains constrained by privacy and data-governance considerations. Anesthesia-specific decision support, particularly for intraoperative hypotension, illustrates both the promise of continuous physiological data streams and the unresolved question of whether earlier prediction reliably changes clinical outcomes rather than simply generating more alerts. We conclude that the principal barrier to wider adoption is no longer algorithmic performance but implementation: interpretability sufficient for clinician trust, governance robust enough for sustained use, and integration tight enough to actually alter scheduling and intraoperative practice.