Computed Tomography-Based Radiomics Model for Predicting the WHO/ISUP Grade of Clear Cell Renal Cell Carcinoma Preoperatively: A Multicenter Study

To examine the ability of computed tomography (CT) radiomic features in multivariate analysis and construction of radiological model for identification of the WHO/ISUP pathological grade of clear cell renal cell carcinoma (CCRCC) tumors.This was a retrospective study design using data from four hospitals from January 2018 to August 2019. There were 197 patients with a definitive diagnosis of clear cell renal cell carcinoma (CCRCC)by pathology or biopsy. These subjects were divided into the training set (n=122) and the validation set (n=75). Enhanced CT lesion scans of two phases (corticomedullary phase, nephrographic phase) of clear cell renal cell carcinoma (CCRCC) were used for whole tumor VOI plots. The IBEX radiomic software package in Matlab was used to extract the radiomic features of tumor VOI images. Next, the Mann-Whitney U test and minimum redundancy-maximum relevance algorithm was used for feature screening and dimensionality reduction. Next, logistic regression combined with Akaike information criterion were used to screen the best prediction model. The performance of the prediction model was tested and compared to independent external validation cohorts. ROC was used to evaluate the discrimination of clear cell renal cell carcinoma (CCRCC) in the training and validation sets.After feature dimensionality reduction and model screening, the logistic prediction model constructed from seven radiomic features showed the best performance in identification of tumor WHO pathological grades. The AUC of the training set was 0.89, and the sensitivity was 0.85. In the independent external validation set, the AUC of the prediction model was 0.81, and the sensitivity was 0.58. The performance of the model was slightly decreased in the independent external validation set but was still maintained at a high level.A radiological model constructed from extracted CT radiomic features can effectively predict the WHO pathological grade of CCRCC tumors; the model shows some clinical generalization. This provides an effective value for patient prognosis and treatment.