Evaluating the Predictive Capability of Radiomics Features of Perirenal Fat in Enhanced CT Images for Staging, and Grading of UTUC Tumours Using Machine Learning

Background: Upper tract urothelial carcinoma (UTUC) often presents with aggressive behaviour, challenging early diagnosis and precise risk stratification. This study explores whether radiomic features extracted from perirenal fat (PRF) around the tumour provide complementary diagnostic and prognostic value beyond conventional tumour-based radiomics. Methods: A retrospective cohort of 103 UTUC patients undergoing radical nephroureterectomy was analysed. Tumour re-gions of interest (ROIs) and concentric PRF expansions (10–30 mm) were segmented from arteri-al-phase CT scans. Radiomic features were extracted using PyRadiomics, filtered by correlation and intraclass correlation coefficients, and integrated with clinical variables (e.g., age, BMI, mul-tifocality). Multiple machines learning models, including MLPClassifier and CatBoost, were evaluated via repeated cross-validation; performance was assessed using area under the ROC curve (AUC), sensitivity, specificity, F1-score, and DeLong tests. Results: The best tumour-grade model (AUC = 0.961) combined tumour and 10 mm PRF features, outperforming PRF-only (AUC = 0.900) and tumour-only (AUC = 0.934) approaches. For tumour stage, a combined model (tumour + 15 mm PRF) achieved AUC = 0.852, surpassing PRF-only configurations (AUC range, 0.711–0.778). DeLong tests indicated that tumour + PRF significantly exceeds PRF-only models, but differences from strong tumour-only baselines were not always significant. Conclusion: PRF ra-diomics contributes complementary predictive information for UTUC grade and stage. These findings underscore the potential of combining tumour and PRF features to refine prognostic models, ultimately supporting more tailored clinical management in UTUC.