Added diagnostic value of pseudocapsule and ADC heterogeneity beyond ccLS v2.0 in small renal masses

To identify which multiparametric MRI parameters provide independent diagnostic value for clear cell renal cell carcinoma (ccRCC) beyond the clear cell likelihood score (ccLS) version 2.0 in small renal masses (SRMs), and to determine whether their addition improves discriminative performance.

This retrospective single-center study included patients with surgically confirmed SRMs (≤ 4 cm, > 25% enhancing component, no macroscopic fat) who underwent preoperative 1.5 T multiparametric MRI between September 2021 and May 2025. Two blinded radiologists of differing experience independently assigned ccLS v2.0 and evaluated qualitative and quantitative MR features. Univariate and multivariate logistic regression identified independent predictors; ROC analysis with the DeLong test assessed the added value of an augmented model over ccLS alone, with calibration and decision-curve analysis. Inter-reader agreement was quantified using Cohen's κ, quadratic-weighted κ, and the intraclass correlation.

Of 817 patients, 108 eligible SRMs (64 ccRCC) across 102 patients were analyzed. At a cutoff of ≥ 4, ccLS v2.0 yielded AUCs of 0.837 (Reader 1) and 0.834 (Reader 2). On multivariate analysis with ccLS entered as a covariate, pseudocapsule and normalized ADC standard deviation (nADCsd) were the independent predictors of ccRCC in both readers, with pseudocapsule the strongest binary predictor (OR 6.96, p = 0.003; OR 4.25, p = 0.015). Adding pseudocapsule and nADCsd to the ccLS score significantly improved discrimination (Reader 1: AUC 0.905 vs. 0.837, p = 0.032; Reader 2: 0.882 vs. 0.834, p = 0.048); after bootstrap optimism correction, the improvement persisted (corrected AUC 0.890 and 0.872), with favorable calibration and net benefit on decision-curve analysis. Within the indeterminate ccLS = 3 subgroup, pseudocapsule was present in 100% and 90% of ccRCC lesions versus 22% and 31% of non-ccRCC lesions, and nADCsd was the only quantitative feature significant in both readers. Pseudocapsule showed excellent inter-reader agreement (κ = 0.862) and nADCsd good agreement (ICC 0.755).

Pseudocapsule presence and nADCsd independently predicted ccRCC beyond ccLS v2.0 and provided significant added value, particularly in indeterminate ccLS = 3 lesions, pending multicenter validation.