The Value Of Radiomics Based On CT Image In The Differential Diagnosis Of Rectal Neuroendocrine Tumors And Rectal Adenocarcinoma

Objective:To investigate the value of CT radiomics in differentiating rectal neuroendocrine tumor(RNET)from rectal adenocarcinoma(READ).Methods:A total of 105 patients who were pathologically diagnosed as rnet(n=49)and read(n=56)in the First Affiliated Hospital of China Medical University from January 2013 to January 2021 were analyzed retrospectively.Each patient underwent rectal enhanced CT scanning.The ROI of lesons was manually sketched on plain scan and venous phase images,and then the corresponding radiomics features were extracted.Principal components analysis(PCA)and ANOVA were used to reduce the dimension of features,select the features with differential diagnosis value,and divide the training set and test set by five fold cross validation test,Create three classifier models:logistic regression(LR),random forest(RF)and decision tree(DT).The area under the curve(AUC),sensitivity and specificity of receiver operating characteristic(ROC)curve were used to evaluate the classification efficiency of the model,and the difference of ROC curve was compared by Delong test.Result:There was no significant difference in age and gender between RNET and READ groups(P>0.05).There was no significant difference in lesion size,plain scan density and enhancement degree between the two diseases(P>0.05);2107 omics features were extracted from each phase of the image,8 features related to differential diagnosis were selected after dimensionality reduction of vein phase omics features,and 6 were selected after dimensionality reduction of plain scan phase omics features.The AUC values of LR,RF and DT classifiers based on venous phase images were 0.89,0.82 and 0.77 respectively.Among them,LR model showed the best classification efficiency,with sensitivity of 0.76 and specificity of 0.75.The AUC difference between LR model and the other two models was statistically significant(P<0.05).Conclusion:The imaging omics model based on CT images is valuable for the differential diagnoss of RNET and READ.However,our results still need to be further verified in a larger cohort.