摘要
目的探讨肺炎人工智能(AI)诊断软件出现假阳性诊断的类型、成因及CT特点。方法选取2020年3月至2020年4月行胸部CT扫描的受检者1061例,分别通过AI诊断软件和2名资深胸部影像诊断医师进行分析检测。结果1061例患者中,AI软件诊断假阳性336例(691灶)。主要为正常/异常解剖结构相关性假灶108灶,约占15.6%;运动相关性假灶97灶,约占14.0%;肺组织含气不足及组织结构异常相关性假灶266灶,约占38.5%;异物相关性假灶7灶,约占1.0%;肺内单位面积内血管密集相关性假病80灶,约占11.6%;多种因素混合相关性假灶133灶,约占19.2%。结论肺炎AI软件存在多种类型的假阳诊断,充分认识其表现,有助于肺炎AI软件在临床工作中发挥最大效益。
Objective To discuss the types,causes and CT characteristics of false positive diagnosis in artificial intelligence(AI)software for pneumonia.Methods One thousand and sixty-one cases of chest CT were enrolled sequentially from March 2020 to April 2020.All the cases were subjected to AI system and two thoracic radiologists.Results A false positive total of 336 cases(691 foci)were detected in 1061 cases,accounting for about 31.7%.It mainly included normal/abnormal anatomic structure-related false foci(108/691,15.6%),exercise-related false foci(97/691,14.0%),false foci associated with deficiency of lung gas and abnormal tissue structure(266/691,38.5%),false foci related foreign body(7/691,1.0%),vascular dense-related pseudodiseases(80/691,11.6%),and false foci mixed multiple factors(133/691,19.2%).Conclusion There are many kinds of false positive diagnosis in AI software for pneumonia,and to fully understand its performance is conducive to the maximum benefit of AI software for pneumonia.
作者
王其军
王倩倩
刘鹏
刘晓亮
马民
刘红光
Wang Qijun;Wang Qianqian;Liu Peng;Liu Xiaoliang;Ma Min;Liu Hongguang(Department of Radiology,Qingdao West Coast New District People′s Hospital,Shandong 266400,China;不详)
出处
《实用医学影像杂志》
2021年第2期115-118,共4页
Journal of Practical Medical Imaging
基金
医学与健康事业研究发展基金项目-伦琴影像科研专项(SD-202008-009)。
关键词
人工智能
假阳性
体层摄影术
X线计算机
Artificial intelligence
False positive
Tomography,X-ray computer
