Renal CT algorithm for evaluating clear cell carcinoma in solid renal masses ≤ 4 cm: a retrospective chinese...

To externally validate a renal CT algorithm for diagnosing clear cell renal cell carcinoma (ccRCC) in solid renal masses ≤ 4 cm and assess inter-reader agreement according to reader experience and across institutions.

This retrospective study included 217 patients (217 masses ≤ 4 cm) from two Chinese centers enrolled between January 2023 and December 2025. Histopathology served as the reference standard for final diagnosis. At each center, three radiologists independently evaluated attenuation ratio, heterogeneity, and CT score. Fleiss' weighted kappa was used to assess inter-reader agreement. Receiver operating characteristic (ROC) curve analysis was performed on original ordinal CT scores to calculate AUC, and sensitivity, specificity, accuracy, positive predictive value (PPV) and negative predictive value (NPV) were calculated at the cutoff CT score ≥ 4.

Center 1 included 121 patients (62 with ccRCC and 59 with non-ccRCC), and Center 2 included 96 patients (48 with ccRCC and 48 with non-ccRCC). Inter-reader agreement was almost perfect for attenuation ratio and CT scores (κ = 0.82-0.85) and substantial for heterogeneity score (κ = 0.72-0.76). Based on ordinal scores, the area under the curve (AUC) ranged from 0.823 to 0.837 (95% CI: 0.753-0.904) in Center 1 and from 0.830 to 0.856 (95% CI: 0.752-0.927) in Center 2, with overall AUC 0.829-0.842 (95% CI: 0.778-0.891). At CT score ≥ 4, sensitivity, specificity, accuracy, PPV, and NPV ranged from 68.8% to 75.0% (95% CI: 53.7%-86.4%), 72.9% to 83.3% (95% CI: 59.7%-92.5%), 71.9% to 79.2% (95% CI: 63.0%-86.8%), 73.3% to 81.8% (95% CI: 60.3%-91.8%), and 70.5% to 76.9% (95% CI: 57.4%-87.5%), respectively. No significant differences in AUC were found among readers with different experience levels (all P > 0.05).

The renal CT score offers favorable specificity and reproducibility for diagnosing ccRCC in small (≤ 4 cm) renal masses. The CT cutoff of ≥ 4 yields stable diagnostic performance across readers and centers. This low-cost algorithm shows promise as a potential tool for primary healthcare settings, pending prospective validation.