Prediction Of Common Subtypes Of Renal Cell Carcinoma Based On T2WI And Dynamic Contrast-enhanced MRI Imaging Radiomics Mode

Objective:The subtype and nuclear grade of renal cell carcinoma are very important for clinical treatment.This study mainly discusses the value of the imaging characteristics obtained from multi-sequence magnetic resonance images in distinguishing clear cell renal cell carcinoma(ccRCC),papillary renal cell carcinoma(pRCC)and chromophobe cell renal cell carcinoma(cRCC).and chromophobe renal cell carcinoma(CRCC).Materials and Methods:From March 2014 to April 2020,84 patients(ccRCC=46 cases,pRCC=20 cases,cRCC=18 cases)confirmed by postoperative pathology and underwent preoperative magnetic resonance imaging were retrospectively studied.3D-Slicer software was used to delineate 3D full-layer Region of interest(ROI)of tumors on three MRI sequences(T2WI,EN-T1WI cortical stage and EN-T1WI medullary stage),and Python software was used to extract image radiomics features from tumor volume.Intra-group inter-group correlation analysis was used to calculate intra-group inter-group correlation coefficient(ICC)for each feature group,and features with ICC>0.75 were selected as stable features that could be extracted repeatedly for further screening.The masses were randomly divided into training sets and validation sets(6:4).Kruskal-wallis test was used to screen out the best texture features for each MRI sequence to identify renal cancer subtypes.P≤0.05 was considered as significant difference,and Countif function is used to screen the feature subset to obtain the final intersection features of the three sequences.Logistic regression models of T2WI,EN-T1WI cortical phase and EN-T1WI medullary phase sequences were established using the selected intersection features.ROC curves for the three subtypes in the three sequences were generated based on radiomic characteristics,and reports verified the AUC,sensitivity and specificity of the three subtypes in the three sequences.Results:A total of 883 characteristics were retained by intra-group and inter-group correlation analysis.There were 16 overlapping imaging features of the three subtypes with significant differences in the three sequences.ROC curve showed significant difference between T2WI,EN-T1WI cortical and EN-T1WI medulla sequences(0.833,0.895 and 0.886 between ccRCC and pRCC;ccRCC and cRCC were 0.822,0.856 and 0.767,respectively.pRCC and cRCC were 0.857,0.881 and 0.857,respectively).Conclusion:Based on image radiomics data obtained from multiple sequence magnetic resonance images,T2WI,EN-T1WI cortical and EN-T1WI medullary imaging models can distinguish ccRCC,pRCC and cRCC well,and en-T1WI cortical imaging models have the best diagnostic performance.