Hopkins Daniel, Callary Stuart A, Solomon L Bogdan, Lee Peter V S, Ackland David C
Department of Biomedical Engineering, University of Melbourne, Parkville, Victoria, Australia.
Centre for Orthopaedic and Trauma Research, University of Adelaide, Adelaide, South Australia, Australia.
J Orthop Res. 2025 Jul;43(7):1315-1324. doi: 10.1002/jor.26086. Epub 2025 May 2.
Revision total hip arthroplasty (rTHA) involving large acetabular defects is associated with high early failure rates, primarily due to cup loosening. Most acetabular defect classification systems used in surgical planning are based on planar radiographs and do not encapsulate three-dimensional geometry and morphology of the acetabular defect. This study aimed to develop an automated computational modeling pipeline for rapid generation of three-dimensional acetabular bone defect geometry. The framework employed artificial neural network segmentation of preoperative pelvic computed tomography (CT) images and statistical shape model generation for defect reconstruction in 60 rTHA patients. Regional acetabular absolute defect volumes (ADV), relative defect volumes (RDV) and defect depths (DD) were calculated and stratified within Paprosky classifications. Defect geometries from the automated modeling pipeline were validated against manually reconstructed models and were found to have a mean dice coefficient of 0.827 and a mean relative volume error of 16.4%. The mean ADV, RDV and DD of classification groups generally increased with defect severity. Except for superior RDV and ADV between 3A and 2A defects, and anterior RDV and DD between 3B and 3A defects, statistically significant differences in ADV, RDV or DD were only found between 3B and 2B-2C defects (p < 0.05). Poor correlations observed between ADV, RDV, and DD within Paprosky classifications suggest that quantitative measures are not unique to each Paprosky grade. The automated modeling tools developed may be useful in surgical planning and computational modeling of rTHA.
涉及大髋臼缺损的翻修全髋关节置换术（rTHA）早期失败率较高，主要原因是髋臼杯松动。手术规划中使用的大多数髋臼缺损分类系统基于平面X线片，未涵盖髋臼缺损的三维几何形状和形态。本研究旨在开发一种自动计算建模流程，用于快速生成三维髋臼骨缺损几何形状。该框架采用术前骨盆计算机断层扫描（CT）图像的人工神经网络分割和统计形状模型生成，对60例rTHA患者的缺损进行重建。计算髋臼区域绝对缺损体积（ADV）、相对缺损体积（RDV）和缺损深度（DD），并在Paprosky分类中进行分层。将自动建模流程生成的缺损几何形状与手动重建模型进行验证，发现平均骰子系数为0.827，平均相对体积误差为16.4%。分类组的平均ADV、RDV和DD一般随缺损严重程度增加。除了3A和2A缺损之间的上方RDV和ADV，以及3B和3A缺损之间的前方RDV和DD外，仅在3B和2B - 2C缺损之间发现ADV、RDV或DD存在统计学显著差异（p < 0.05）。在Paprosky分类中观察到的ADV、RDV和DD之间的相关性较差，表明定量测量并非每个Paprosky分级所特有。所开发的自动建模工具可能有助于rTHA的手术规划和计算建模。
