Li Qiang, Yang Wenzhuo, Xu Meng, An Nan, Wang Dawei, Wang Xing, Jin Hui, Wang Jiajiong, Wang Jincheng
Department of Orthopedics, The Second Hospital of Jilin University, 218 Ziqiang Street, Changchun 130021, Jilin, People's Republic of China.
Institute of Advanced Research, Infervision Medical Technology Co., Ltd, People's Republic of China.
Biomed Phys Eng Express. 2021 Apr 14;7(3). doi: 10.1088/2057-1976/abf483.
Developmental dysplasia of the hip (DDH) is a common orthopedic disease. A simple and cost-effective scientific tool for assisting the early diagnosis of DDH is urgently needed. This study proposed a new artificial intelligence (AI) model for automated measure of the CE angle to aid the diagnosis of DDH by modifying the Mask R-CNN algorithm.13228 anteroposterior pelvic x-ray images were collected from the PACS system of the second Hospital of Jilin University, of which 104 images were randomly selected as test data. The rest of x-ray images were labelled and preprocessed for model development. The new AI model was the constructed based modified Mask R-CNN model to detect key points for CE angle measurement. The performance of AI model on measuring CE angle was verified by comparing with three attending orthopaedic doctors. The mean CE angles on left and right pelvis measured by the AI model was 29.46 ± 6.98°and 27.92 ± 6.56°, respectively, while the mean CE angle measured by the three doctors was 29.85 ± 6.92°and 27.75 ± 6.45°, respectively. AI model displayed a higly consistency with the doctors in measuring CE angles. Besides, AI model showed a much high efficiency in term of measuring time-consumption. In this study, we successfully constructed a new effective model for measuring CE angle by identifying key points, which provided a new intelligent measurement tool for orthopedic image measurement and evaluation.
发育性髋关节发育不良（DDH）是一种常见的骨科疾病。迫切需要一种简单且经济高效的科学工具来辅助DDH的早期诊断。本研究通过改进Mask R-CNN算法，提出了一种用于自动测量CE角以辅助DDH诊断的新型人工智能（AI）模型。从吉林大学第二医院的PACS系统中收集了13228张骨盆前后位X线图像，其中随机选择104张图像作为测试数据。其余X线图像进行标注和预处理以用于模型开发。基于改进的Mask R-CNN模型构建新的AI模型，以检测CE角测量的关键点。通过与三名骨科主治医生比较，验证了AI模型在测量CE角方面的性能。AI模型测量的左右骨盆平均CE角分别为29.46±6.98°和27.92±6.56°，而三名医生测量的平均CE角分别为29.85±6.92°和27.75±6.45°。AI模型在测量CE角方面与医生表现出高度一致性。此外，AI模型在测量耗时方面显示出更高的效率。在本研究中，我们通过识别关键点成功构建了一种新的有效测量CE角的模型，为骨科图像测量和评估提供了一种新的智能测量工具。
