AI与不同年资医师对肺结节检出效能的比较研究  

Comparative study on the detection efficiency of pulmonary nodules between AI and radiographers with different seniority

作　　者：于程程 王晨 李翔 刘东宇 张贺诚 YU Chengcheng;WANG Chen;LI Xiang;LIU Dongyu;ZHANG Hecheng(Department of Radiation,Beijing Hospital of Integrated Traditional Chinese and Western Medicine,Beijing 100039,China)

机构地区：[1]北京中西医结合医院放射科,北京100039

出　　处：《影像研究与医学应用》2025年第4期22-25,共4页Journal of Imaging Research and Medical Applications

摘　　要：目的:探讨基于深度学习的人工智能(AI)与不同级别医师CT肺结节检测能力的差异。方法:随机抽取2022年3月—2023年9月于北京中西医结合医院行胸部CT平扫的患者110例,以1名从事胸部影像诊断的副主任医师及1名高年资主治医师共同阅片检出肺结节的数目、位置、大小为参照标准,比较A组(2名住院医师)、B组(2名中低年资主治医师)、C组(AI)对肺结节的检出率、假阳性率。结果:副主任医师及高年资主治医师共确定1320枚结节。A、B、C组检出率分别为60.9%、74.5%和81.9%;假阳性率分别为15.5%、13.4%和13.1%。三组微小、中等大小结节、实性、亚实性、钙化密度、胸膜下、血管旁、其他位置的结节检出率比较,差异有统计学意义(P<0.05);但三组大结节检出率比较,差异无统计学意义(P>0.05)。结论:AI能有效地检出肺结节,尤其是在微小结节、亚实性和钙化结节检出率上高于中低年资主治医师,可作为肺结节筛检的有效辅助工具。Objective To investigate the differences in CT lung nodule detection capabilities between artificial intelligence(AI)based on deep learning and radiographers of various levels.Methods A total of 110 patients who underwent chest CT scans at Beijing Hospital of Integrated Traditional Chinese and Western Medicine between March 2022 and September 2023 were randomly selected.The number,location,and size of lung nodules detected by a deputy chief radiographer specializing in chest imaging diagnosis and a senior attending radiographer were used as the reference standard.The detection rates and false positive rates of lung nodules were compared among Group A(two resident radiographers),Group B(two intermediate-to low-seniority attending radiographers),and Group C(AI).Results The deputy chief radiographer and the senior attending radiographer identified a total of 1320 nodules.The detection rates for Groups A,B,and C were 60.9%,74.5%,and 81.9%,respectively;the false positive rates were 15.5%,13.4%,and 13.1%,respectively.There were statistically significant differences in the detection rates of micro-nodules,medium-sized nodules,solid nodules,subsolid nodules,calcified nodules,nodules located subpleurally,adjacent to blood vessels,and in other locations among the three groups(P<0.05).However,there was no statistically significant difference in the detection rate of large nodules among the three groups(P>0.05).Conclusion AI can effectively detect lung nodules,particularly micro-nodules,subsolid nodules,and calcified nodules,with higher detection rates than intermediate-to low-seniority attending radiographers.AI can serve as an effective auxiliary tool for lung nodule screening.

关 键 词：肺结节 人工智能 CT 胸部CT 对照研究 

分 类 号：R445.3[医药卫生—影像医学与核医学]