Nguyen Thong Phi, Chae Dong-Sik, Park Sung-Jun, Kang Kyung-Yil, Lee Woo-Suk, Yoon Jonghun
Department of Mechanical Design Engineering, Hanyang University, 222, Wangsimni-ro, Seongdongsu, Seoul, 04763, Republic of Korea.
Department of Orthopedic Surgery, International St. Mary's Hospital, Catholic Kwandong University College of Medicine, Incheon, Republic of Korea.
Comput Biol Med. 2020 May;120:103732. doi: 10.1016/j.compbiomed.2020.103732. Epub 2020 Mar 29.
One of the first tasks in osteotomy and arthroplasty is to identify the lower limb varus and valgus deformity status. The measurement of a set of angles to determine this status is generally performed manually with the measurement accuracy depending heavily on the experience of the person performing the measurements. This study proposes a method for calculating the required angles in lower limb radiographic (X-ray) images supported by the convolutional neural network. To achieved high accuracy in the measuring process, not only is a decentralized deep learning algorithm, including two orders for the radiographic, utilized, but also a training dataset is built based on the geometric knowledge related to the deformity correction principles. The developed algorithm performance is compared with standard references consisting of manually measured values provided by doctors in 80 radiographic images exhibiting an impressively low deviation of less than 1.5° in 82.3% of the cases.
截骨术和关节成形术的首要任务之一是确定下肢内翻和外翻畸形状态。通常通过手动测量一组角度来确定这种状态，测量精度在很大程度上取决于测量者的经验。本研究提出了一种由卷积神经网络支持的计算下肢X线图像中所需角度的方法。为了在测量过程中实现高精度，不仅使用了一种去中心化深度学习算法，包括两个用于X线摄影的指令，还基于与畸形矫正原理相关的几何知识构建了一个训练数据集。将所开发算法的性能与由医生手动测量值组成的标准参考进行比较，在80张X线图像中，82.3%的病例显示出令人印象深刻的低偏差，偏差小于1.5°。
