Abstract: BACKGROUND: There is currently a lack of extensive multi-center clinical research on the accuracy of the artificial intelligence (AI) HIP planning system for predicting the prosthesis size. Compared with the two-dimensional template planning, it remains poorly understood whether it is more accurate and whether it has advantages in restoring the acetabular and femoral eccentricity and rotation center. 
OBJECTIVE: To investigate the accuracy of AI HIP preoperative planning system in predicting the size of total hip arthroplasty prosthesis and its influencing factors, and to analyze its value in restoring eccentricity and rotation center. 
METHODS: According to the random number table method, 70 cases were divided into two groups (n=35). AIHIP (study group) and Smart Joint two-dimensional digital template (control group) were used for preoperative planning. The accuracy of predicting the prosthesis size between the two groups was compared, and the potential factors affecting the accuracy of AI HIP prediction were analyzed, including age, gender, body mass index, International Society of Osseous Circulation (ARCO) staging and femoral medullary cavity classification (Dorr classification). The acetabular eccentricity, femoral eccentricity, horizontal and vertical displacement difference of the center of rotation, and the length difference of both lower extremities were measured before and after operation.   
RESULTS AND CONCLUSION: (1) The prediction of the prosthesis size on the acetabular side and femoral side was 57.1% (20/35) and 42.9% (15/35) in the study group, and 31.4% (11/35) and 34.3% (12/35) in the control group. The prosthesis compliance rates of the study group on acetabular side and femoral side were 94.3% (33/35) and 82.9% (29/35), while those of the control group were 74.3% (26/35) and 62.9% (22/35); the difference between the two groups was statistically significant (P < 0.05). (2) Logistic regression analysis showed that gender, body mass index and femoral canal classification had no effect on the accuracy of the femoral stem. Compared with the control group, AI HIP had more advantages in planning the Dorr A femoral medullary cavity, and the comparison between the two groups was statistically significant (P=0.012). The postoperative femoral eccentricity and total eccentricity in the control group were greater than those in the study group, and the differences were statistically significant (all P < 0.05). (3) In the horizontal direction, the average displacement of the center of rotation was 1.8 (-0.7, 4.1) mm in the study group, and 2.3 (-0.5, 5.0) mm in the control group; in the vertical direction, the average displacement in the study group was 0.8 (-1.5, 2.7) mm and -1.5 (-3.6, 2.5) mm in the control group, with no statistical significance (P > 0.05). There was no significant difference in the length of both lower limbs between the two groups after operation, which was 3.05 (2.01, 3.89) mm in the study group and 3.63 (2.60, 5.30) mm in the control group (Z=-1.736, P=0.083). (4) These results confirm that the AI HIP preoperative planning system has higher accuracy than the digital two-dimensional template in predicting the size of the acetabular and femoral stem prostheses, and has more advantages in planning the shape of the Dorr A femoral medullary canal. Using artificial intelligence deep learning technology helps to more reliably predict the length and eccentricity of the lower limbs and restore the center of rotation of the hip joint.

Key words: total hip arthroplasty, artificial intelligence, two-dimensional planning, preoperative planning, template measurement