423 Current Applications and Future Directions of Artificial Intelligence (AI) and Machine Learning in Spine Surgery. A Scoping Review

Abstract

INTRODUCTION: AI and machine learning simulate human intelligence, enabling computers to learn and make data-based decisions without explicit programming. In fields like spine surgery, where robotic-assisted procedures are growing rapidly, the synergies between these technologies present numerous unexplored applications. METHODS: A scoping review was performed systematically on PubMed to assess current and possible future use of machine learning in spine surgery. The variables of interest were accuracy of screw insertion, pre-operative planning with or without three-dimensional reconstruction and outcome prediction with or without surgical recommendations. RESULTS: A total of 2771 articles were found within the primary query. 59 studies were selected for final review. The findings were structured around the key advantages that AI/machine learning currently offers for spine surgery. One important application is AI's integration with robotic-assisted screw placement, with recent studies showing up to 97% accuracy, even for historically difficult screws like S2AI. Neuronavigational software advancements have facilitated AI integration in surgical planning. Machine learning is increasingly used for predictive analytics in treatment decision-making, achieving up to 83% accuracy compared to ground truth. AI's ability to assess the 3D relationship between anatomical structures and surgical channels demonstrates high inter-rater reliability of 0.96 with ground truth, as scored by intraclass correlation coefficient (ICC). A recent application involves analyzing surgical behaviors, useful for tasks like resident training and identifying time-consuming or risky surgical stages based on action frequency. CONCLUSIONS: Few AI/machine learning applications have been explored in spine surgery, but the potential for various uses—from training to surgical planning—is vast and growing annually. With advancements in self-learning software and robotic capabilities, a future where decision-making software integrates with highly mobile robots may be on the horizon.