基于卷积神经网络的X线片下肢关节角度识别算法  

Lower limb joint angle calculation algorithm based on convolutional neural network in X-ray films

作　　者：刘静妮 盛玉武 赵长秀 牛存良 黄国源 许长栋 赵姗姗 陈彬 LIU Jingni;SHENG Yuwu;ZHAO Changxiu;NIU Cuniang;HUANG Guoyuan;XU Changdong;ZHAO Shanshan;CHEN Bin(Department of Radiology,Wuwei People's Hospital,Wuwei 733000,China;Department of Orthopedics,Wuwei People's Hospital,Wuwei 733000,China)

机构地区：[1]武威市人民医院放射科,甘肃武威733000 [2]武威市人民医院骨科,甘肃武威733000

出　　处：《中国医学物理学杂志》2024年第8期996-999,共4页Chinese Journal of Medical Physics

基　　金：甘肃省武威市科技计划项目(WW23B02SF056)。

摘　　要：提出一种基于卷积神经网络的X线片下肢关节角度识别算法,首先在X线片中使用Yolov5目标检测模型来识别特定类别的感兴趣区域,并使用U-Net模型进行热图回归来识别关键特征点,最后进行下肢关节角度的计算。研究结果表明,本文提出的算法相比于之前的算法精度更高,结果准确可靠,为临床研究和实践提供参考。A convolutional neural network-based algorithm is proposed for calculating lower limb joint angle in X-ray films.After identifying the region of interest of a specific category in X-ray films through Yolov5 object detection model,U-Net model is used to perform heat map regression for identifying the key feature points,and then the lower limb joint angle is calculated.The results show that the proposed algorithm has higher accuracy than the previous algorithms and can obtain accurate and reliable results,providing references for clinical research and practice.

关 键 词：卷积神经网络 目标检测 特征点定位 下肢力线 

分 类 号：R318[医药卫生—生物医学工程] TP18[医药卫生—基础医学]