基于CheXNet辅助诊断系统在肺部常见病诊断中的应用研究  

Application Study of CheXNet-Based Assisted Diagnosis System in Diagnosis of Common Lung Diseases

作　　者：窦瑞欣[1] 黄鹏 白永明 张晓曦[1] 李琼[1] DOU Ruixin;HUANG Peng;BAI Yongming;ZHANG Xiaoxi;LI Qiong(Department of Radiology,Tianjin Nankai Hospital)

机构地区：[1]天津市南开医院放射科

出　　处：《中国医学计算机成像杂志》2021年第6期505-509,共5页Chinese Computed Medical Imaging

摘　　要：目的:研究基于CheXNet辅助诊断系统在肺部常见病诊断中的应用价值和临床意义。方法:利用CheXNet结合神经网络输出可视化技术Grad-CAM生成预测热图,对我院2018年1月—2019年9月经临床诊断的100例肺部常见病(肺部肿瘤50例,肺部感染50例)患者的胸部X线检查进行辅助诊断。结果:以临床确诊为金标准,在CheXNet的辅助下检出肺部肿瘤48例、肺部感染47例;直接读片检出肺部肿瘤41例、肺部感染39例。直接读片与CheXNet辅助下的检出率存在差异且有统计学意义。在CheXNet辅助下诊断准确度(96.5%)优于直接读片(80%),差异具有统计学意义(P<0.05)。结论:借助CheXNet不仅可减轻放射诊断的工作量,还能提高肺部常见病的检出率及诊断准确度,尤其是床旁胸部X线检查对危重症患者行的影像学辅助诊断,更有助于临床进行快速和有效的治疗。Purpose: To study the application value and clinical significance of CheXNet-based assisted diagnosis system in the diagnosis of common lung diseases.Methods: By using CheXNet combined with neural network output visualization technology Grad-CAM to generate predictive heat maps, the chest X-ray examination of 100 patients with common lung diseases(50 cases of lung tumors and 50 cases of lung infections) clinically diagnosed in our hospital from January 2018 to September 2019 was used for assisted diagnosis.Results: With the clinical diagnosis as the gold standard, 48 cases of lung tumors and 47 cases of lung infections were detected with the assistance of CheXNet, and 41 cases of lung tumors and 39 cases of lung infections were detected by direct reading. There was a significant difference in the detection rate between direct reading and CheXNet assistance, which was with statistical significance. With the help of CheXNet, the diagnostic accuracy(96.5%) was better than that by direct reading(80%), and the difference was with statistical significance(P<0.05). Conclusion: The application of CheXNet not only reduces the workload of radiological diagnosis, but also improves the detection rate and diagnostic accuracy of common lung diseases, especially for the assisted diagnosis of chest X-ray examination of critically ill patients,which is more helpful for rapid and effective clinical treatment.

关 键 词：神经网络 深度学习 人工智能 辅助诊断 胸部X线 CheXNet 

分 类 号：R445.2[医药卫生—影像医学与核医学]