Development and validation of a preoperative MRI habitat radiomics model to predict variant histology in bladder cancer

Motivation: Bladder cancer (BCa) patients with variant histology (VH) exhibit more aggressive tumor characteristics, resulting in poor prognosis and resistance to neoadjuvant therapy. Accurate assessment of VH is challenging due to biological complexity and limited biopsy samples. Goal(s): To identify VH in BCa and predict patient prognosis. Approach: An MRI-based fusion model that combines habitat radiomics, clinicopathologic variables, and conventional radiomics was developed, using data from 764 bladder cancer patients across four institutions. Results: The proposed model demonstrated high predictive accuracy (AUC 0.895-0.974) across multiple cohorts, with low-risk patients showing better disease-free survival and higher pathological complete response rates. Impact: The proposed VHRiS model represents a promising and robust tool forto identifying variant histologyVH, offering providing valuable insights into patient prognosis within the microtumoral environment and suggesting indicating potential benefit for precision clinical decision-making guidance in patients with bladder cancer.