CT的形态学征象联合影像组学特征诊断模型在肺结节诊断中的应用  

Application of the CT morphological signs combined with radiomics features nomogram for the diagnosis of solid pulmonary nodules

作　　者：刘冬冬 武志峰[2] 鄂林宁[2] 王荣华 LIU Dongdong;WU Zhifeng;E Linning;WANG Ronghua(Department of Medical Imaging,Shanxi Medical University,Taiyuan 030001,P.R.China;CT Division,Shanxi Academy of Medical Science,Shanxi Bethune Hospital,Taiyuan 030032,P.R.China)

机构地区：[1]山西医科大学医学影像学系,山西太原030001 [2]山西医学科学院山西白求恩医院CT室,山西太原030032

出　　处：《医学影像学杂志》2021年第2期229-233,共5页Journal of Medical Imaging

基　　金：山西省自然科学基金支持项目(编号:201801D121200,201701D121151)。

摘　　要：目的结合肺结节CT形态学征像与影像组学特征建立联合诊断模型,期望提高孤立性肺结节的诊断准确性。方法选取2012年2月~2019年1月206例肺实性结节(直径≤2cm)的CT图像。由两名医师独立进行形态学征像评估,并诊断结节良恶性,使用PyRadiomics开源软件提取、筛选定量影像组学特征,采用逻辑回归建立影像组学模型,并融合形态学征像建立联合诊断模型,采用受试者操作特征(ROC)曲线评价模型诊断效能,并与人工诊断结果进行比较。结果结节的血管集束、支气管截断、毛刺、分叶、胸膜牵拉、空泡征及充气支气管征在恶性结节中的检出率高,与良性结节之间的差异有统计学意义(P<0.05);结节越小,各种CT征像检出率越低。两位医师对肺结节征像识别一致性存在差异,高年资医师诊断效能明显高于低年资医师,曲线下面积(AUC)分别为0.81(95%CI,0.744~0.856),0.69(95%CI,0.623~0.753)。影像组学诊断模型对良恶性肺结节具有较高的鉴别诊断效能,AUC为0.86(95%CI,0.8~0.92),结合毛刺征、分叶征、胸膜牵拉征及影像组学特征建立联合诊断模型,进一步提高肺结节诊断效能,AUC为0.92(95%CI,0.87~0.96)。结论结合肺结节CT形态学征像及影像组学特征建立的联合诊断模型,在肺结节良恶性鉴别诊断中有较高的应用价值。Objective To establish a combined diagnosis model of solitary pulmonary nodules by combining CT morphological signs with radiomics feature,in order to improve the accuracy of diagnosis of solitary pulmonary nodules with diameter≤2 cm.Methods The CT images of 206 cases of solid pulmonary nodules with diameter≤2 cm from 2012 to 2019 were analyzed retrospectively.Morphologic signs were independently evaluated by two radiologists,and benign and malignant nodules were diagnosed at the same time.Radiomics features were extracted and screened by the open source software of Pyradiomics.The radiomics model was established by logistic regression,and the combined diagnosis model was established by combining morphological signs with radiomics features.The ROC curve was used to evaluate the diagnostic efficiency of the model,and the results were compared with the manual diagnostic performance.Results The detection rate of vascular convergence,bronchial truncation,spiculation,lobulation,pleura retraction,vacuole and air bronchogramsign in malignant nodules was higher than that in benign ones,with statistical significance(P<0.05).The detection rate of CT signs decreased with the reduce of nodule volume.There was a difference between the two doctors in the recognition of CT morphological signs,the diagnostic efficacy of the senior doctors was significantly higher than that of the junior doctors,with AUC of 0.81(95%CI,0.744~0.856)and 0.69(95%CI,0.623~0.753),respectively.The radiomic diagnosis model was more effective in the differential diagnosis of benign and malignant pulmonary nodules,the AUC was 0.86(95%CI,0.8~0.92).Finally,a combined diagnosis model was established by combining spiculation sign,lobulation sign and pleura traction sign with radiomic features.The AUC was 0.92(95%CI,0.87~0.96).Conclusion The combined diagnosis model based on the CT morphological and radiomic features of pulmonary nodules has a high application value in the differential diagnosis of benign and malignant pulmonary nodules.

关 键 词：肺结节 体层摄影术 X线计算机 影像组学 列线图 

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