Xiong et al. developed deep learning models to assess the malignancy and aggressiveness of renal masses using preoperative computed tomography (CT) scans. The study analyzed 13,261 CT volumes from 4,557 patients, creating two convolutional neural networks: one predicting malignancy and another differentiating between aggressive and indolent tumors. The malignancy model achieved an area under the curve (AUC) of 0.871 in a prospective test set, outperforming six out of seven experienced radiologists. The aggressiveness model attained an AUC of 0.783, surpassing conventional radiomics-based models and the nephrometry score nomogram in accuracy. AI predictions correlated with patient survival, showing significantly worse disease-specific survival (DSS), recurrence-free survival (RFS), and overall survival (OS) for AI-predicted aggressive tumors, with hazard ratios as high as 20.61 in external validation. The models were tested across multiple cohorts, including an independent external test set and a prospective validation set, demonstrating consistent performance. AI-based aggressiveness scores were found to be independent prognostic factors, outperforming TNM staging and ISUP grading in predicting survival. Genetic and immune profiling revealed aggressive tumors had increased AHNAK2 mutations and a highly infiltrated yet immunosuppressive tumor microenvironment, with elevated CD8+ T cells and Tregs, while indolent tumors had higher mast cell infiltration.