Objective:

To investigate the value of radiological models in differenting small renal cell carcinoma (diameter 4 cm) from fat-poor renal angiomyolipoma (fpAML) through three-phase enhanced computed tomography (CT) radiomic features.

Methods:

A total of 125 cases with renal tumors with diameter 4 c

m were retrospectively studied

which including small renal carcinoma（

n

=80）and fpAML（

n

=45）. The region of interest（ROI） was delineated on the images of the corticomedullary phase

the nephrographic phase and the excretion phase

respectively. Then the training set (

n

=93) and the testing set (

n

=32) were divided according to the ratio of 3∶1. After radiomics features were extracted and screened

support vector machine (SVM) model and logistic regression (LR) model were established respectively. using receiver operating characteristic (ROC) curve evaluation model for differential diagnosis of small renal cell carcinoma and fpAML. ROC curve was used to evaluate the effectiveness of model in differential diagnosis of small renal cell carcinoma and fpAML.

Results:

A total of 91 optimal features were selected by dimensionality reduction

39 in corticomedullary phase

25 in nephrographic phase and 27 in excretion phase. The radiological model based on the image of corticomedullary phase has the best discriminative efficacy

and the area under curve (AUC)

specificity

sensitivity and accuracy of the training set were 0.961

0.917

0.939

0.882

of the training set were 0.825

0.900

0.750 and 0.875

respectively.

Conclusion:

Three phase enhanced CT adiological models have a strong effectiveness in the differential diagnosis of small renal cell carcinoma and fpAML

in which the differential efficiency of the corticomedullary phase is better than the nephrographic phase and excretion phase

and the differential value of SVM model is higher than LR model.