Abstract 5779: A comparison of CT/RNASeq and MRI/RNASeq-based radiogenomics for staging muscle-invasive bladder cancer: a pilot study

Abstract Introduction: Accurate staging of bladder cancer (BCa) assists in identifying optimal treatment (e.g., transurethral resection vs. radical cystectomy vs. bladder preservation). Current clinical tumor staging methods for BCa leverages axial computed tomography (CT) scans due to its speed and cost-effectiveness. Prior research has demonstrated staging uncertainty with CT, leading to under- or over-staging of patients. Improving staging accuracy is crucial for treatment decisions. Recent work has suggested that MRI can better emphasize contrast between abnormal structures and thereby improve stratification, while genomics signatures can identify molecular changes underpinning tumor spread (e.g., collagen remodeling). Thus, we hypothesize that magnetic resonance imaging (MRI)/RNASeq-based radiogenomics can more accurately stage BCa. Method: A total of 31 matched CT scans, MRI and formalin-fixed paraffin-embedded (FFPE) tissues were collected and made available for analysis. CT and MR images were reviewed by at least two radiologists to generate a consensus BCa T-stage. Next, RNASeq gene expression was profiled from TURBT FFPE tissue using the nf-rnaseq workflow to generate count matrices from the reads. PyRadiomics was used to extract MR/CT features. A radiogenomic analysis was conducted to identify imaging signatures of advanced T-stage through integration of MR/CT imaging and RNASeq data, using Bayesian LASSO for feature selection followed by unpenalized regression modeling, adjusted for age and sex. The performance of image-only and radiogenomic signatures were compared between MR/CT through likelihood ratio testing (LRT). Results: CT and MRI resulted in the same accuracy of staging pT2 or less. Interestingly, the accuracy of staging >pT2 is better in MR than CT scan, suggesting that MR resolution is superior to CT allowing more accurate staging. CT radiogenomic signatures (p=5.9e-4) stratified patients more effectively than the MR radiogenomics (p=2.9e-3) though changes appeared to be driven by genomic features more than imaging features. In reviewing hierarchical clustering of the radiogenomic features (Fig. 1), we found that MR imaging influenced clustering of the signature more than in CT data, speculating that the MR data delivers more useful information other than tumor stage (e.g. predicting survival and treatment response in addition to stage). Conclusion: These early findings from the pilot study suggest promising results. The limited sample size restricts our ability to draw definitive conclusions or apply more sophisticated machine learning methods, such as the investigation of how genes may further stratify radiomic features by disease progression and vice versa. Citation Format: Joshua Levy, Toru Sakatani, Kaoru Murakami, Takashi Kobayashi, Charles J. Rosser, Hideki Furuya. A comparison of CT/RNASeq and MRI/RNASeq-based radiogenomics for staging muscle-invasive bladder cancer: a pilot study [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 5779.