825 Background: Urothelial bladder cancer exhibits substantial pathological, molecular, and clinical heterogeneity. Genomic and transcriptomic profiling of muscle-invasive bladder cancer (MIBC) has identified recurrent alterations that inform classification and therapeutic targets. However, evolving mutational landscapes and the infrastructure required for testing limit routine clinical applicability. Computed tomography (CT), used routinely for staging and surveillance, may offer a noninvasive method to infer tumor biology. Radiomics - the extraction of quantitative imaging features - may help link phenotypic imaging signatures to underlying molecular alterations. Methods: We integrated genomics data from The Cancer Genome Atlas with CT data from The Cancer Imaging Archive in 89 patients with biopsy-proven MIBC to create models for prediction of DNA mutations, Tumor Mutational Burden (TMB), and mRNA expression. An in-house developed CT-based radiomics pipeline was used to compute 488 texture metrics for the segmented images and quantify visual characteristics such as brightness distribution, pixel relationships, and periodic structural patterns. Three machine learning classifiers - Random Forest, Extreme Gradient Boosting, and Elastic Net - were trained on the radiomics data and evaluated with 10-fold cross-validation using area under the receiver-operator curve (AUC) as a balanced performance measure. Results: Among 15 DNA mutations found in at least 10% of the cohort, EP300, FGFR3, and ARID1A were predicted most reliably, with AUCs of 0.77 and 0.76, and 0.75 respectively. The models also predicted tumors with high TMB (AUC = 0.61), transcriptomic patterns associated with poor prognosis by two mRNA panels (AUC = 0.73, AUC = 0.65), and transcriptional levels of key cell cycle (CDKN1A, AUC = 0.78) and apoptotic (CASP3, AUC = 0.71) genes. Finally, the model could frequently discriminate the luminal infiltrated molecular subtype from other variants (AUC = 0.69). Conclusions: Our study demonstrates that CT-derived radiomics features can capture biologically and clinically relevant information in muscle-invasive bladder cancer. These findings support the potential utility of radiomics as a noninvasive, scalable adjunct to genomic profiling in MIBC.
Boyne et al. (Sun,) studied this question.