Comparative Analysis of CT and MRI Combined with RNA Sequencing for Radiogenomic Staging of Bladder Cancer

TL;DR

This study compares CT and MRI imaging with RNA sequencing to improve bladder cancer staging, finding MRI more accurate and suggesting potential for radiogenomic approaches.

Contribution

The study introduces a radiogenomic approach combining MRI radiomics and RNASeq to enhance bladder cancer staging accuracy.

Findings

MRI was more accurate than CT in staging tumors beyond T2.

MRI radiogenomic signatures outperformed CT in staging patients.

Radiomic features were linked to genomic signatures of advanced tumor stages.

Abstract

Accurate staging of bladder cancer (BCa) is important for identifying optimal treatment. Currently, clinical tumor staging for BCa relies on computed tomography (CT) scans, but these can lead to under- or overstaging of patients. Recent research suggests that using magnetic resonance imaging (MRI) along with RNA sequencing (RNASeq) gene expression analysis can provide more precise staging. In this study, 31 matched CT scans, MRI images, and formalin-fixed, paraffin-embedded (FFPE) tissues were collected. First, two radiologists reviewed the images for staging BCa. Next, radiomics features were extracted from both CT and MR images, and computational radiogenomics analyses were performed. Subsequently, RNASeq was performed using FFPE tissues of TURBT prior to cystectomy. A radiogenomic analysis was conducted to identify advanced T-stage signatures. Regarding imaging alone, MRI was found to be more accurate in staging >T2 compared to CT scans. Within a retrospective cohort, MRI radiogenomic signatures were more effective in staging patients than CT, with genomic features playing a significant role. Using canonical correlation analysis, we additionally identified radiomic features underlying genomic signatures of advanced tumor stage. When applying these signatures across a small prospective cohort, MRI radiomic data were able to stratify stage; however, the addition of the same genomic features did not improve the sensitivity and specificity of the model. These preliminary results are promising, but additional research with larger sample sizes is needed to draw definitive conclusions and explore further correlations and statistical interactions between genes and imaging features through machine learning techniques as we move radiogenomics to the clinic.

Figures