Objective To explore the value of CT features combined with texture analysis in differentiating pulmonary scle-rosing pneumocytoma(PSP)from peripheral lung cancer(PLC).Methods 136 PSP patients and 131 PLC patients confirmed in the First Affiliated Hospital of Zhengzhou University were collected.All cases were randomly divided into training set and veri-fication set according to the ratio of 7:3.The imaging features of two groups of CT images and the texture of CT thin-layer images in venous phase were analyzed.The image features and texture parameters with statistical differences were used to construct mul-tivariate binary logistic regression model.And then,we drew ROC curve,calculated AUC value,and evaluated and compared the effectiveness of each model in diagnosing PSP and PLC.Results There were statistical differences in morphology,calcifica-tion,liquefaction necrosis,lobulation,burr,pleural indentation,cavity,mediastinal/hilar lymph node enlargement and the best texture parameters of 16 groups between the two groups.The feature model of CT image consisted of lobulation,burr,pleu-ral indentation and mediastinal/hilar lymph node enlargement.The five optimal texture parameters that included in the regression model of training set were Perc.10％,S(3,0)SumAverg,S(4,0)SumAverg,S(4,4)SumAverg and WavEnLH s-1.In the train-ing set,the AUC of PSP and PLC diagnosed by CT image feature model and CT texture parameter model were 0.847 and 0.851,respectively,and there was no significant difference between them(P = 0.912).The AUC of imaging features combined with tex-ture parameter model was the highest,which was 0.939,and its accuracy,sensitivity and specificity were 85.0％,82.1％and 93.5％respectively.In the verification set,the AUC of image features combined with texture parameter model in diagnosing two groups of lesions was 0.923,which was higher than that of CT image feature model(AUC=0.864;;Z=2.627,P=0.009)and CT texture parameter model(AUC=0.832;;Z=2.147,P=0.031).Conclusion CT image combined with texture analysis has good diagnostic value for distinguishing PSP from PLC.
维普
万方数据