Chunhui, LiAl Mopti, Abdulrahman2025-07-092025-05https://hdl.handle.net/20.500.14154/75779This thesis investigates the diagnostic and prognostic potential of perirenal fat (PRF) radiomics in upper urinary tract cancers through three interconnected studies. Using computational techniques to extract quantitative features from CT images, the research establishes PRF as a valuable biomarker for tumour behaviour assessment. The first study, examining clear cell renal cell carcinoma in 474 patients, demonstrates that models integrating tumour features, PRF radiomics, and clinical variables achieve high accuracy for tumour grade (AUC 0.780) and stage prediction (AUC 0.829). Analysis reveals that PRF regions at 4-10mm radial distances from tumours contain the most predictive information. The second study on upper tract urothelial carcinoma (UTUC) reveals excellent performance of combined models in predicting tumour grade (AUC 0.961) and stage (AUC 0.852). PRF-only models also show substantial discriminative capability, confirming that PRF contains distinct textural patterns associated with tumour aggressiveness. The final study establishes the prognostic value of PRF radiomics in UTUC through survival analysis, with the combined radiomics-clinical model achieving a C-index of 0.784. Key radiomics features emerge as strong prognostic indicators, particularly when integrated with clinical variables like stage and hydronephrosis. Methodologically, the research develops a semi-automated approach for PRF analysis and implements a standardised radiomics workflow. The findings contribute novel insights by establishing PRF as an independent source of diagnostic information, developing standardised methodology for analysis, identifying specific radiomic signatures of aggressive disease, and creating predictive models that outperform conventional assessment. This work demonstrates that non-invasive analysis of PRF can enhance risk stratification and treatment planning in upper urinary tract cancers.153enCTRCCPERIRENAL FATThesis