CT-Based Radiomics for Prediction of Molecular Markers in Clear Cell Renal Cell Carcinoma: A Comprehensive Review

Background and Objectives: Clear cell renal cell carcinoma (ccRCC) demonstrates substantial molecular and clinical heterogeneity, limiting the prognostic accuracy of conventional staging system and complicating treatment selection. CT-based radiomics and radiogenomics have emerged as promising non-invasive approaches for predicting molecular biomarkers. This review aimed to evaluate the current evidence regarding CT-based radiogenomics for the prediction of molecular markers in ccRCC, with emphasis on methodological approaches, predictive performance, and clinical applicability. Materials and Methods: A comprehensive literature search of PubMed/MEDLINE, Scopus, and Cochrane Library databases was performed for original studies published between January 2012 and December 2025. Eligible studies included patients with histopathologically confirmed ccRCC, performed CT-based radiomics feature extraction, and investigated molecular or genetic biomarkers using machine learning (ML) methods. Data regarding CT acquisition phase, segmentation strategy, radiomics features, ML algorithms, investigated biomarkers, and model performance metrics were extracted. Results and Discussion: Twenty-five retrospective studies were included. CT-based radiomics demonstrated promising performance in predicting gene mutations, including Von Hippel-Lindau (VHL), Polybromo 1 (PBRM1), BRCA1-associated protein 1 (BAP1), SET domain containing 2 (SETD2), and Lysine demethylase 5C (KDM5C), with reported area under the curve (AUC) values reaching 0.987. Radiogenomic models also showed utility in assessing hypoxia-related pathways, lipid metabolism signatures, programmed cell death profiles, immune-related markers, and tumor microenvironment characteristics, including programmed death-ligand 1 (PD-L1), Cluster of Differentiation 68 (CD68+) tumor-associated macrophages (TAMs), Cytotoxic T-Lymphocyte-Associated Protein 4 (CTLA-4), Forkhead Box P3 (FOXP3), and Ki-67 proliferation index. Predictive performance varied across biomarkers, with AUCs generally ranging from 0.68 to 0.91. Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), Adaptive Boosting (AdaBoost), and Gradient Boosting algorithms were most commonly applied. Conclusions: CT-based radiogenomics represents a promising non-invasive tool for molecular characterization and risk stratification in ccRCC. Standardized multicenter prospective studies, methodological homogeneity, and external validation are required before routine clinical implementation.