Intraoperative navigation using augmented reality technology and artificial intelligence in surgery

Objective.To summarize the clinical experience of using augmented reality for preoperative planning and intraoperative navigation across several surgical specialties. Material and methods. From 2022 through 2025, 74 AR-assisted procedures were performed at the University Clinic of the Russian University of Medicine, Ministry of Health of the Russian Federation. The series included procedures in abdominal, interventional, maxillofacial, reconstructive, head and neck, and urologic surgery. Results. Intraoperative setup of the AR system required no more than 15 min and did not materially increase the overall procedure time. Among patients with head and neck tumors, mean operating time was 110±10 min with AR guidance, compared with 130±10 min for comparable procedures performed without AR. Three-dimensional visualization influenced the choice of surgical approach in 5 of 14 patients (35.7%). No perioperative complications were recorded in this group. During transarterial chemoembolization, AR facilitated spatial interpretation of the vascular anatomy. Procedure and fluoroscopy times tended to be shorter, although the differences between groups did not reach statistical significance. Post-procedure Likert-scale ratings provided by the surgical teams supported the system’s usability and its perceived contribution to improved visualization and navigational accuracy. Conclusion. These findings support the clinical feasibility of AR navigation and suggest that it may be particularly useful in anatomically complex regions, in the presence of variant vascular anatomy, and when deep structures are poorly visualized. However, the heterogeneity of the study population and the absence of standardized outcome measures preclude regarding this series as definitive evidence of effectiveness. Further advances will require more accurate segmentation and spatial registration, compensation for intraoperative tissue deformation, and automation of routine model-preparation steps. The authors’ approach to the real-time accumulation of spatially registered ultrasound data represents a promising avenue for addressing some of these limitations.