Comparison of artificial intelligence-assisted or CT-based 3-dimensional preoperative planning with conventional 2-dimensional templating in total hip arthroplasty

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Background: Implant planning is essential for prosthesis matching in total hip arthroplasty (THA). Conventional 2-dimensional (2D) templating has limitations in predicting implant size. Recently, artificial intelligence (AI)-assisted 3-dimensional (3D) planning and computed tomography (CT)-based 3D planning have been increasingly applied clinically. However, 3D planning methods remain heterogeneous, making the evidence difficult to interpret. This study aimed to systematically evaluate AI-assisted or CT-based 3D preoperative planning in THA compared with conventional 2D templating. Primary outcomes included implant size prediction accuracy, operative time, intraoperative blood loss, and postoperative limb length discrepancy (LLD). Methods: A systematic literature search was conducted according to PRISMA guidelines across 7 databases from inception to August 2025 to identify clinical studies comparing AI-assisted or CT-based 3D planning with conventional 2D templating. Outcomes included exact-match implant size prediction accuracy, accuracy within 1 size, operative time, intraoperative blood loss, and postoperative LLD. Risk of bias was assessed using RoB 2 for randomized trials and ROBINS-I for non-randomized studies. Evidence certainty was assessed using GRADE. Fixed-effect or random-effects models were selected by heterogeneity. Results: Twelve studies were included. Because some studies used paired 2D and 3D templating comparisons within the same patients, and some outcomes were reported at the component level, pooled analyses were performed separately by outcome. Compared with 2D templating, AI-assisted or CT-based 3D planning showed higher exact-match prediction accuracy for acetabular cups and femoral stems, with odds ratios of 4.50 and 4.54, respectively. Accuracy within 1 size also favored 3D planning (OR = 5.17). The 3D planning group had shorter operative time (mean difference [MD] = −17.26 minutes), less intraoperative blood loss (MD = −32.66 mL), and smaller postoperative LLD (MD = −1.64 mm). However, operative time heterogeneity was substantial (I 2 = 92%), and some analyses suggested small-study effects. Conclusion: AI-assisted or CT-based 3D preoperative planning may improve implant size prediction accuracy and selected perioperative outcomes compared with conventional 2D templating. However, because the included 3D planning methods were heterogeneous and not all systems explicitly incorporated AI algorithms, these findings support 3D planning strategies rather than the independent superiority of AI itself. High-quality randomized trials are needed.