Abstract: BACKGROUND: Artificial intelligence planning system can automatically establish a three-dimensional model and generate planning schemes, but its accuracy in predicting the prosthesis size has not been fully verified. 
OBJECTIVE: To investigate the accuracy of artificial intelligence planning system in predicting prosthesis size before total hip arthroplasty and its influence on clinical prognosis, and further analyze the risk factors affecting the accuracy of planning. 
METHODS: Clinical data of patients with unilateral initial total hip arthroplasty who were admitted to the Department of Orthopedics of Affiliated Hospital of Xuzhou Medical University from January 2021 to June 2022 were prospectively collected. The patients were randomly divided into the artificial intelligence planning system group (n=80) and the conventional template group (n=79). Intraoperative use of prostheses and preoperative planning of prosthesis matching were compared between the two groups. Postoperative follow-up Harris scores and the occurrence of complications such as leg length discrepancy, dislocation and prosthesis loosening were recorded in both groups. The effects of demographic indicators, preoperative diagnosis, and Dorr typing on the accuracy of femoral stem planning were explored using univariate and multivariate Logistic regression analyses.
RESULTS AND CONCLUSION: (1) The prediction of the prosthesis size on the acetabular side and femoral side was 50%(40/80) and 55%(44/80) in the artificial intelligence planning system group, compared to 34%(27/79) and 37%(29/79) in the conventional template group, with statistically significant differences (P < 0.05). (2) The artificial intelligence planning system group had an accuracy rate within one size difference for the acetabular and femoral side prostheses of 91%(73/80) and 86%(69/80), compared to 82%(65/79) and 72%(58/79) in the conventional template group, with differences statistically different only on the femoral side (P < 0.05). (3) No dislocation or prosthesis loosening occurred in the two groups during postoperative follow-up. The difference in lower limb length between the artificial intelligence planning system and conventional template groups was (3.56±2.32) mm and (3.52±2.41) mm. At the last follow-up, the Harris scores of the artificial intelligence planning system and conventional template groups were (92.74±3.08) and (91.81±3.52), respectively; there was no significant difference in the above differences (P > 0.05). (4) Univariate analysis results showed that preoperative diagnosis as developmental dysplasia of the hip and osteonecrosis of the femoral head, and Dorr type B and C femurs had a significant effect on the accuracy of predicted prosthesis size using an artificial intelligence planning system (P < 0.05). (5) Multivariate logistic regression analysis showed that preoperative diagnosis of developmental dysplasia of the hip (OR=18.233, 95%CI: 2.662-124.888) was an independent risk factor for the prediction of femoral stem size by artificial intelligence planning system. (6) The artificial intelligence planning system has a higher accuracy in predicting prosthetic size than traditional two-dimensional templates, and there is not a significant difference in the risk of postoperative complications or joint function. The accuracy of the artificial intelligence planning system in patients with developmental dysplasia of the hip was low due to anatomical deformities and acetabular anatomical position reconstruction.

Key words: preoperative planning, artificial intelligence, total hip arthroplasty, proximal femoral canal bone morphology, prosthesis size