基于深度学习的DR胸片智能质控方法研究  

Research on Intelligent Quality Control Method of DR Chest Film Based on Deep Learning

作　　者：王平[1] 胡博奇[1] 安东洪[1] 刘蓄蕾 石张镇[1] 田中生 郝富德 刘景鑫[1] WANG Ping;HU Boqi;AN Donghong;LIU Xulei;SHI Zhangzhen;TIAN Zhongsheng;HAO Fude;LIU Jingxin(Department of Radiology,China-Japan Union Hospital,Jilin University,Changchun Jilin 130021,China;WX Medical Technology Co.,Ltd.,Shenyang Liaoning 110000,China)

机构地区：[1]吉林大学中日联谊医院放射科,吉林长春130033 [2]辽宁万象联合医疗科技有限公司,辽宁沈阳110000

出　　处：《中国医疗设备》2020年第10期28-33,共6页China Medical Devices

基　　金：国家重点研发计划项目(2018YFC0116901,2018YFC1315600)。

摘　　要：目的为了减轻医学影像质量控制工作给医院及相关机构带来的工作负担,弥补人工质量控制存在的多方面缺陷,本文提出一种基于深度学习的DR胸片正位和侧位摄影的质量控制方法。方法通过特定结构的卷积神经网络完成DR图像的语义分析,结合医学影像质量控制标准,自动批量完成影像质控工作。结果该方法在多个数据集中的平均准确率为98.32%,单例影像的平均质控时间为83 ms。结论该模型可以快速、准确地实现DR影像的自动智能质控。Objective In order to reduce the burden of medical image quality control work on hospitals and related institutions, and make up for the various defects in manual quality control, this paper proposes a quality control method based on deep learning for DR radiography of Chest. Methods The semantic analysis of DR images was completed through a convolutional neural network with a specific structure, combined with medical image quality control standards, to automatically complete image quality control in batches. Results The average accuracy of the method in multiple data sets was 98.32%, and the average quality control time of a single image was 83 ms. Conclusion The model can realize automatic DR image quality control quickly and accurately.

关 键 词：深度学习 胸部DR 质量控制 卷积神经网络 

分 类 号：R540.41[医药卫生—心血管疾病]