CT Texture Analysis In Differentiating Different Histopathological Subtype Of Renal Cell Carcinoma

Objectives:(1)To evaluate the accuracy 3D of texture analysis(TA)in differentiation of different pathological type of renal cell carinoma with plain scan CT images and triphasic contrast-enhanced CT images.(2)To evaluate the diagnostic efficacy of symbiotic matrix related texture parameters in the diagnosis of renal cell carcinoma subtypes.Methods:The clinical data of 67 patients with renal cell carcinoma were retrospectively analyzed in our study,including 35 cases of clear cell renal cell carcinoma,14 cases of cheomophobe renal cell carcinoma,18 cases of papillary renal cell carcinoma from September 2016 to December 2018.They were confirmed by histopathological examination.All patients had abdominal renal CT scan with four phases(precontrast phase,corticomedullary phase,nephrographic phase and excretory phase).Two high-grade urological disease imaging diagnosticians readed all the phases of CT images and screened out images containing all levels of the lesion.If there are different opinions,try to reach consensus.The non-enhanced and contrast-enhanced CT three-dimensional structural images of the lesions were manually delineated by MaZda software,and extracted texture features from manually regions of interest.The feature selection methods mutual information(MI),classification error probability combined with average correlation coefficients(POE+ACC),Fishers coefficient and the combination of the above three methods(MPF)were included.The statistical methods including linear discriminant analysis(LD A),nonlinear discriminant analysis(NDA),raw data analysis(RDA)and principal component analysis(PCA)were used to distinguish the subtypes of renal cell carcinoma.The results were shown by misclassification rate.Results:(1)In the CT scan with four phases,the texture features for differentiating the subtypes of renal cell carcinoma were mainly from corticomedullary phase which had the lowest misclassification rate 0.00%to 8.96%.(2)Fisher feature selection combined with NDA statistical method based on corticomedullary phase which had the lowest misclassification rate 0.00%.The area under the ROC curve(AUC)values ranged from 0.952 to 0.990,with sensitivities from 87.5%to 100.00%and specificities from 88.57%to 97.14%.(3)Enhanced scan cortical symbiotic matrix parameters S(0,1,0)DiffEntrp?S(0,1,0)Contrast?S(1,1,0)DiffEntrp?S(0,1,0)DifVarnc?S(1,-1,0)DifEntrp?S(1,-1,0)Contrast?S(0,1,0)Correlat?S(1,-1,0)Correlat?S(1,1,0)Correlat?S(0,1,0)IDM the area under the curve of differential diagnosis of renal cell carcinoma pathological subtypes(AUC)is between 0.952 and 0.990,The sensitivity and specificity were 87.5%and 100%,88.5%and 88%,respectively.Between 57%and 97.14%,the accuracy is 89.55%?95.52%,which can be used to identify and diagnose the pathological subtypes of renal cell carcinoma.Conclusions:(1)3D texture analysis based on CT images might be a reliable quantitative method for differentiating the subtypes of renal cell carcinoma,especially on corticomedullary phase images.(2)The correlation texture parameters of 10 symbiotic matrices extracted by renal CT enhanced scan included S(0,1,0)DiffEntrp?S(0,1,0)Contrast?S(1,1,0)DifEntrp?S(0,1,0)DifVarnc?S(1,-1,0)DifEntrp?S(1,-1,0)Contrast?S(0,1,0)Correlat?S(1,-1,0)Correlat?S(1,1,0)Correlat?S(0,1,0)IDM can be used to distinguish pathological subtypes of renal cell carcinoma.