基于AI对肺小结节CT图像层厚与结节特征的检出效能影响  

The effect of image thickness and nodule feature on the performance of pulmonary small nodules using CT-Based AI

作　　者：周围 胡富碧[1] 刘亚斌[1] 李书平[1] 吴少平[1] ZHOU Wei;HU Fubi;LIU Yabin;LI Shuping;WU Shaoping(Department of Radiology,The First Affiliated Hospital of Chengdu Medical College,Chengdu 610500,P.R.China)

机构地区：[1]成都医学院第一附属医院放射科,四川成都610500

出　　处：《医学影像学杂志》2021年第2期234-238,共5页Journal of Medical Imaging

摘　　要：目的探讨基于深度学习的人工智能(AI)辅助诊断系统对肺小结节的CT检出效能,并观察不同图像层厚与结节特征对其检出效能的影响。方法选取我院胸部CT平扫检查200例,导入人工智能软件工作站中,根据不同CT图像层厚、结节类型、结节大小以及机型,分别记录AI与医师对结节的检出情况,比较其检出率、灵敏度、假阳性率以及检出时间,并比较AI及医师的检出情况。结果AI在厚层(5 mm)与薄层(1.5 mm)对于pGGN的检出差异无统计学意义(P>0.05);AI在厚层对于mGGN的检出高于薄层(t=2.282,P=0.025);AI在薄层对于SN的检出明显高于厚层(t=-10.377,P<0.001),AI在薄层对pGGN、mGGN及SN检出灵敏度均明显高于厚层,经统计学分析分别为,t=-4.823,P<0.001,t=-4.048,P<0.001,t=-10.186,P<0.001。AI在64排CT机型及16排CT机型中共检出肺小结节分别为491、627枚,t=-0.428,P=0.427,P>0.05,二者差异无统计学意义。结论CT扫描层厚越厚,基于深度学习的AI对肺小结节检出的漏诊率越高,CT扫描层厚越薄,AI对肺小结节的检出越有利,但其假阳性率和检出时间也随之增高。Objective To investigate the detection efficiency of artificial intelligence-assisted diagnosis system based on deep learning on CT pulmonary nodules,and to study the effects of different image layer thickness and nodule characteristics on its detection efficiency.Methods 200 cases of chest CT scans in our hospital were collected and imported into an artificial intelligence software workstation.According to different CT image layer thickness,nodule type,nodule size and model,the detection of nodules by AI and physicians was recorded separately.We compared its detection rate,sensitivity,false positive rate and detection time,and compare the detection of AI and physicians.Results AI in the thick layer(5 mm)and thin layer(1.5 mm)was not statistically significant for the detection of pGGN(P>0.05);AI in the thick layer was higher than the thin layer in the detection of mGGN(t=2.282,P=0.025);AI detection of SN in thin layer was significantly higher than that of thick layer(t=-10.377,P<0.001),AI detection sensitivity of pGGN,mGGN and SN in thin layer was significantly higher than thick layer.The analysis was t=-4.823,P<0.001,t=-4.048,P<0.001,t=-10.186,P<0.001.In the 64-row CT model and the 16-row CT model,AI detected a total of 491 and 627 pulmonary nodules,t=-0.428,P=0.427,P>0.05,the difference between the two was not statistically significant.Conclusion The thicker the CT scan layer thickness,the higher the rate of missed diagnosis of small pulmonary nodules based on deep learning artificial intelligence AI.The thinner the CT scan layer thickness,the better the detection of AI on small pulmonary nodules,but its false positive rate and detection time also increase.

关 键 词：肺小结节 人工智能 检出率 体层摄影术 X线计算机 

分 类 号：R734.2[医药卫生—肿瘤] R814.42[医药卫生—临床医学]