Accuracy of CT-Based Radiomics Models for Preoperative Grading of Clear Cell Renal Cell Carcinoma: A Systematic Review and Meta-analysis.

Purpose: High-grade clear cell renal cell carcinoma (ccRCC) is linked to aggressive behavior, higher metastatic potential, and poor overall prognosis. This study aims to assess the diagnostic performance of CT-based radiomics models for preoperative differentiation of high-grade from low-grade ccRCC.

Methods: A comprehensive systematic search was conducted across PubMed, Scopus, and Embase. Study quality was evaluated using the QUADAS-2 and METRICS tools. Pooled estimates of sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and area under the curve (AUC) were calculated using a bivariate random-effects model. Separate comparative analyses were performed based on tumor grading system, imaging phase, and model type. Subgroup analysis was performed to identify potential sources of heterogeneity. Sensitivity analysis was conducted to identify potential outliers. Publication bias was assessed using Deek's funnel plot asymmetry test.

Results: A total of 49 studies, encompassing 11,125 patients and 11,922 lesions, were included in the review. Of these, 27 studies were eligible for meta-analysis. The pooled performance of the radiomics model demonstrated a sensitivity of 0.79, specificity of 0.85, PLR of 4.6, NLR of 0.30, DOR of 19.21, and AUC of 0.88. Pooled sensitivity exhibited substantial heterogeneity (I²=51.4%), while heterogeneity for other estimates was minimal. Studies using WHO/ISUP tumor grading system showed higher pooled sensitivity compared to those using Fuhrman grading system (Sensitivity: 0.81 vs 0.71, p=0.03, AUC: 0.88 vs. 0.82, p<0.001). Combined models performed comparably to radiomics-only models (AUC: 0.87 vs. 0.85, p=0.1). Nephrographic phase (AUC: 0.67) showed inferior performance compared to corticomedullary phase (AUC: 0.82) and non-contrast images (AUC: 0.83). No considerable publication bias was observed (p=0.73).

Conclusion: CT-based radiomics models have demonstrated promise as non-invasive tools for grading ccRCC and may serve as valuable adjuncts in clinical practice.