Accuracy assessment of an artificial intelligence model in predicting intraoperative bone resection parameters for robotic-assisted total knee arthroplasty

The purpose of this study is to preliminarily evaluate the performance of our AI-assisted decision-making system compared to the final intraoperative parameters obtained under the supervision of the operating surgeon. This prospective study included 52 patients (54 knees) undergoing primary robotic-assisted TKA from November 2024 to February 2025. The AI system, developed from retrospective data, incorporated 17 parameters including resection depths, angles and gap measurements. We compared initial osteotomy parameters, AI predictions, and final implemented parameters. Accuracy was categorized as high (0–1 mm/°), medium (1–2 mm/°), or low precision (> 2 mm/°). The AI system achieved high precision across all parameters. For femoral varus/valgus angle, 100.00% of predictions fell with high precision, with 59.26% exactly matching final resection. Medial distal femoral resection showed 87.04% high precision. Similar high precision rates were observed for lateral distal femoral resection (83.33%), femoral rotation (88.89%), posterior condylar resections (98.15% medial, 92.59% lateral), and tibial resections (83.33% medial, 94.44% lateral and 88.89% for tibia varus/valgus angle). The AI decision support system demonstrated remarkable accuracy in predicting bone resection parameters for robotic-assisted TKA. These findings support AI integration in preoperative planning to enhance surgical precision and efficiency.