Musbahi Omar, Hadjixenophontos Savvas, Gill Saran S, Soteriou Iris, Pouris Kyriacos, Ueno Takuro, Cobb Justin P
MSk Lab, White City Campus, Imperial College London, London, UK.
Department of Electrical and Electronic Engineering, Imperial College London, London, UK.
Arthroplast Today. 2025 Jun 3;33:101717. doi: 10.1016/j.artd.2025.101717. eCollection 2025 Jun.
Radiographic assessment is crucial for the success of a hip arthroplasty procedure as a correctly positioned prosthesis indicates favorable long-term outcomes. This project aims to develop a novel artificial intelligence (AI)-based method that can (1) automatically identify the presence of a hip resurfacing prosthesis in radiographs and (2) calculate the radiographic neck-shaft angle (NSA) of the prosthesis from 2-dimensional plane images using both anterior-posterior (AP) and lateral radiographs with high accuracy.
Using a computer vision and pattern recognition algorithm, the femur shaft and prosthesis regions were identified, and their respective angles were extracted for NSA calculation. A neural network (NN) was then trained using clinician-generated AP radiograph NSAs as ground truths and AI-generated AP and lateral NSAs as features. Spearman's correlation and Kruskal-Wallis tests were calculated to explore any significant association between the final AI-generated and clinician-generated AP radiographic NSAs. Mean absolute error (MAE) and R-squared values were calculated with and without the NN model to identify the model's accuracy and variability.
There was a statistically significant correlation between the final AI-generated AP radiographic NSAs and the clinician-generated AP radiographic NSAs (r = 0.93, < .01). MAE, R, and r without the NN were 3.09, 0.37, and 0.83 ( < .01), respectively. MAE and R with the NN were 1.94 and 0.53, respectively.
This study demonstrates that the identification of hip resurfacing prostheses using AI is feasible. By incorporating additional features such as the lateral NSA, the model can provide an accurate prediction of the AP radiographic NSA, closely approximating the ground truth.
影像学评估对于髋关节置换手术的成功至关重要，因为假体位置正确预示着良好的长期预后。本项目旨在开发一种基于人工智能（AI）的新方法，该方法能够（1）在X线片中自动识别髋关节表面置换假体的存在，以及（2）使用前后位（AP）和侧位X线片从二维平面图像中高精度计算假体的影像学颈干角（NSA）。
使用计算机视觉和模式识别算法，识别股骨干和假体区域，并提取它们各自的角度以计算NSA。然后使用临床医生生成的AP X线片NSA作为真值，以及AI生成的AP和侧位NSA作为特征来训练神经网络（NN）。计算Spearman相关性和Kruskal-Wallis检验，以探索最终AI生成的和临床医生生成的AP影像学NSA之间的任何显著关联。在有和没有NN模型的情况下计算平均绝对误差（MAE）和R平方值，以确定模型的准确性和变异性。
最终AI生成的AP影像学NSA与临床医生生成的AP影像学NSA之间存在统计学显著相关性（r = 0.93，P <.01）。没有NN时的MAE、R和r分别为3.09、0.37和0.83（P <.01）。有NN时的MAE和R分别为1.94和0.53。
本研究表明使用AI识别髋关节表面置换假体是可行的。通过纳入诸如侧位NSA等附加特征，该模型可以提供对AP影像学NSA的准确预测，非常接近真值。
