双能CT融合图像在人工智能肺结节筛查中检测效能的探索研究

Investigation on the Detection Capability of Artificial Intelligence for Pulmonary Nodules Using Dual-energy Fusions Image

目的探索双能CT融合120 kVp图像在人工智能(Artificial Intelligence,AI)辅助诊断软件肺结节筛查中的检测效能。方法收集我院行双源CT肺结节筛查的患者381例,随机分为单源扫描(A组)和双能扫描(B组)两组。A组183例,管电压为120 kVp;B组198例,管电压为双源100/Sn 140 kVp,两组均采用CareDose 4D技术。记录容积CT剂量指数(CTDIvol)值、剂量长度乘积(Dose-Length Product,DLP)值。使用AI软件对图像进行结节检测,记录结节的大小、位置、类型(实性、亚实性)。统计AI软件检测全部结节数、不同大小结节数(≥4 mm、<4 mm)、不同类型结节数(实性、亚实性),并结合金标准得到相应的真阳数、假阳数、漏诊数,计算相应的结节检测敏感度、精确度和假阳性率,比较两组间的差异,P<0.05为差异具有统计学意义。结果B组图像的肺结节检测敏感度高于A组(P<0.05),辐射剂量低于A组(P<0.05),同时,B组肺结节检测总的假阳性率低于A组。结论双能融合120 kVp图像在AI软件肺结节检测中的检测效能优于单源扫描120 kVp图像,其辐射剂量更低,更加适用于AI软件肺癌筛查。

Objective This study explores the pulmonary nodules detection capability of an artificial intelligence diagnostic system using dual-energy CT fusion with 120 kVp image.Methods 381 lung cancer screening patients underwent dual-energy CT scans in our hospital were prospectively enrolled in this study and randomly divided into two groups:group A(120 kVp single energy scan,183 cases)and group B(100/Sn 140 kVp dual-energy scan,198 cases).Both groups were treated with CareDose 4D technology.Volumetric CT Dose index(CTDIvol)value and dose-length Product(DLP)value were recorded.AI software was used to detect the nodules in the image,and the size,location and type of nodules(solid and subsolid)were recorded.By comparing with the golden standard,total number of detected nodules was calculated,as well as nodules with different size(≥4 mm and<4 mm),nodules with different density(solid and subsolid),and TPF,FPF,FNF.Finally,the sensitivity,precision and false positive rate of the AI system were obtained.Difference comparision studies were conducted,and P<0.05 for the difference was statistically significant.Results The sensitivity of pulmonary nodules detection in group B was higher than that in group A(P<0.05),and the radiation dose was lower than that in group A(P<0.05).Meanwhile,the total false positive rate of pulmonary nodules detection in group B was lower than that in group A.Conclusion It is concluded from this study that dual-energy fusion 120 kVp image is more effective than single-source scan 120 kVp image in detection of lung nodules,and its radiation dose is lower,which is more suitable for detection of lung cancer by AI software.