Abstract
Although renal pelvic and ureteral urothelial carcinoma share similarities in their origins, disparities on a genetic and clinical level make them divergent entities. Clinical information from the Surveillance, Epidemiology, and End Results (SEER) database was used to validate the characteristics and molecular subtypes using single-center data, which were compared between the two types of muscle-invasive tumors. Simultaneously, to expand the sample size for further verification, we explored a deep learning algorithm to correctly classify molecular subtypes from H&E histology slides. We suggested that the renal pelvic group might have a proclivity towards luminal and the ureter towards basal and P53-like. Furthermore, we explore the heterogeneity of matrix and immune tumor microenvironment, and the ureteral group had more immune cell infiltration and higher stiffness. Collectively, these results showed that muscle-invasive upper tract urothelial carcinoma exist in distinct properties of clinical characteristics, molecular subtype, and tumor microenvironment.