Robotic-referenced automated measurement of acetabular cup orientation on postoperative CT: development and internal validation.
Authors
Affiliations (7)
Affiliations (7)
- Department of Bone and Joint Surgery, Second Affiliated Hospital of Xi'an Jiaotong University, Xi'an, China.
- School of Automation, Xi'an University of Posts and Telecommunications, Xi'an, China.
- Honghui Hospital, Xi'an Jiaotong University, Xi'an, China.
- Department of Orthopaedics, Peking University Third Hospital, Beijing, China.
- Department of Bone and Joint Surgery, Second Affiliated Hospital of Xi'an Jiaotong University, Xi'an, China. [email protected].
- School of Automation, Xi'an University of Posts and Telecommunications, Xi'an, China. [email protected].
- Department of Bone and Joint Surgery, Second Affiliated Hospital of Xi'an Jiaotong University, Xi'an, China. [email protected].
Abstract
Accurate measurement of acetabular cup orientation after total hip arthroplasty is essential, but postoperative CT relies on subjective, error-prone manual measurement. Robot-assisted surgery provides intraoperative cup-orientation measurements but is costly and not widely available. We aimed to develop and evaluate a deep-learning model measuring cup orientation on postoperative CT, using robotic navigation values as reference standard. This secondary analysis of a randomized trial (ChiCTR2200060115) analyzed 94 hips with robotic intraoperative angle measurements and postoperative CT (May 2023 to May 2024). A VGG16-based U-Net segmented key anatomical structures into three-dimensional point clouds. Three measurement pathways were compared: manual annotation, machine learning, and deep learning. The optimal model was integrated into a graphical user interface. Ninety-four hips (mean age, 57.0 years ± 9.5 [standard deviation]; 62 men) were evaluated using 27,821 CT images. The best model (PointNet++) achieved mean absolute errors of 4.48° (anteversion) and 3.89° (inclination), compared with 4.08°/ 5.52° for machine learning and 8.91°/ 8.70° for manual measurement, significantly outperforming manual measurement (both p < 0.05). Exploratory full-cohort Lewinnek classification was correct in 81/94 hips versus 57/94 with manual measurement; in internal validation, 71% and 81% of predictions were within 5° of the robotic reference. A deep-learning model developed using robotic navigation values showed promising internal-validation performance for automated cup-orientation measurement on postoperative CT; independent external validation is required before broader clinical application.Trial registration ChiCTR2200060115, 19 May 2022.