To predict the World Health Organization (WHO)/International Society of Urological Pathology (ISUP) grading of clear cell renal cell carcinoma (ccRCC) using multiple machine learning models based on contrast-enhanced CT, and to compare their performance, thereby exploring a non-invasive method for preoperative pathological grading. Seventy-six patients with ccRCC (65 low-grade and 11 high-grade) from September 2021 to August 2023 were retrospectively included. Regions of interest (ROI) were manually delineated on enhanced CT images, and radiomics features were extracted. The dataset was first split into training and test sets (50%/50%). To address class imbalance, the training set was oversampled using the RandomOverSampler algorithm. Feature scaling and selection using the Maximum Information Coefficient (MIC) were applied strictly within the training set. Radiomics models were built using twelve machine learning classifiers. Diagnostic performance was evaluated using the receiver operating characteristic (ROC) curve, area under the curve (AUC), and decision curve analysis (DCA). Among the 12 classifiers, five (GradientBoostingClassifier, RandomForestClassifier, AdaBoostClassifier, XGBClassifier, and CatBoostClassifier) reported an AUC of 1.000 on the test set; XGBClassifier had the highest sensitivity (0.944) and accuracy (0.969). The high AUC values should be interpreted with caution due to the small sample size. DCA suggested that most models provided a net benefit across a wide threshold range. Machine learning models based on contrast-enhanced CT show potential as a non-invasive tool for distinguishing high-grade from low-grade ccRCC. This preliminary study suggests a promising direction; however, further validation with larger, multi-center cohorts is required before clinical application.
Li et al. (Thu,) studied this question.