基于深度学习的桡骨远端骨折自动分型研究  

Research on Automatic Classification of Distal Radius Fractures Based on Deep Learning

作　　者：杨锋 丛日坤 王卫国 丁波[1] Yang Feng;Cong Rikun;Wang Weiguo;Ding Bo(Network Information Center of Shandong University of Traditional Chinese Medicine,Jinan,Shandong 250355,China;The First Clinical College of Shandong University of Traditional Chinese Medicine,Jinan,Shandong 250355,China)

机构地区：[1]山东中医药大学网络信息中心,山东济南250355 [2]山东中医药大学第一临床医学院,山东济南250355

出　　处：《激光与光电子学进展》2021年第12期216-224,共9页Laser & Optoelectronics Progress

基　　金：山东省社科规划研究项目(19CSHJ13);教育部高等教育司产学合作协同育人项目(201902205004)。

摘　　要：为解决桡骨远端骨折内部病灶区域骨碎块多且不规则,致使医生漏诊及误诊率高的问题,利用课题组前期收集的临床桡骨远端骨折病例设计了一种监督式桡骨远端骨折自动分型的深度学习诊断模型。实验中还引入迁移学习思想,提高了诊断模型的训练效率。最后采用交叉验证的方法对模型进行评估,结果表明,本文提出诊断模型的分类结果优于传统机器学习及经典深度学习分类模型,分类准确率达到了84.2%,较经典深度学习模型提升了4%左右,且网络结构简单,运算速度快,具有一定鲁棒性和较强的泛化能力。In order to solve the problem that there are many and irregular bone fragments in the focal area of the distal radius fracture,which causes the doctor’s missed diagnosis and high rate of misdiagnosis,this paper uses the clinical cases of distal radius fracture collected by the research group to propose a supervised automatic distal radius fracture deep learning model.The experiment also introduces the concept of migration learning,which improves the training efficiency of the diagnostic model.Finally,the experiment uses a cross-validation method to evaluate the model.The results show that the classification results of the proposed diagnostic model are better than traditional machine learning and classic deep learning classification models.The classification accuracy rate reaches 84.2%,which is 4% higher than the classic deep learning model.The network structure is simple,the calculation speed is fast,with certain robustness and strong generalization ability.

关 键 词：图像处理 桡骨远端骨折 预处理 深度学习 迁移学习 过拟合 

分 类 号：O436[机械工程—光学工程]