Artificial intelligence-assisted three-dimensional preoperative planning with patient-specific cutting guides versus conventional planning and instrumentation in total knee arthroplasty: a prospective exploratory cohort study

Abstract Background Total knee arthroplasty (TKA) outcomes may be influenced by preoperative planning and accurate execution. This study compared an AI-assisted three-dimensional preoperative planning and patient-specific instrumentation (PSI) workflow with conventional planning and instrumentation for primary unilateral TKA. The AI-assisted workflow was a CT-based preoperative software platform for repeated virtual osteotomy, implant selection, component-position review, and virtual postoperative joint motion-state visualization; it was not robotic execution or a validated predictor of postoperative range of motion. Methods In this single-center prospective exploratory cohort, 22 patients with knee osteoarthritis undergoing primary unilateral TKA were non-randomly allocated to an AI-assisted planning/PSI group (n = 11) or a Conventional group (n = 11). Outcomes were assessed preoperatively and at 2 and 6 months postoperatively. Clinical, perioperative, radiographic, and gait outcomes were compared using standard between-group and repeated-measures analyses. P values were interpreted as nominal because of the small sample size and multiple comparisons. Results At 6 months, WOMAC scores were lower in the AI-assisted planning/PSI group than in the Conventional group [15.00 (14.00, 17.00) vs. 17.00 (15.00, 18.00); nominal P = 0.02]. KSS function scores were higher in the AI-assisted planning/PSI group at 2 and 6 months [80.00 (70.00, 90.00) vs. 55.00 (55.00, 60.00), and 80.00 (80.00, 90.00) vs. 70.00 (70.00, 70.00); both nominal P < 0.001]. Pain-related outcomes, HALP, drainage volume, and gait parameters showed directional between-group differences that were interpreted as exploratory and descriptive. Postoperative HKA deviation and HKA outlier rates were comparable, while coronal tibial component error was smaller in the AI-assisted planning/PSI group [0.4 (0.2, 0.7) vs. 0.9 (0.6, 1.1) degrees; nominal P = 0.018]. Conclusions In this exploratory cohort, AI-assisted planning with PSI was associated with selected short-term functional, perioperative, gait, and radiographic component-positioning differences, but not superior overall HKA alignment. Findings are preliminary and require confirmation in larger randomized or well-matched studies.