AI and Cancer Research ── Bladder ── 2026-10-31
Bladder cancer: treatment, trials and AI
- Abstract
Muscle-invasive bladder cancer (MIBC) frequently develops resistance to platinum-based chemotherapy, and tools for characterizing treatment-associated tumor microenvironmental features remain limited. We developed an artificial intelligence (AI)-based framework integrating pathological imaging, multiomics data, single-cell RNA sequencing, and ex vivo drug-sensitivity testing. Three convolutional neural network models trained using hematoxylin-and-eosin-stained whole-slide images from The Cancer Genome Atlas (TCGA) classified tumor, necrosis, and lymphocyte-enriched regions and were evaluated in an independent tissue microarray cohort. Patches negative for all three classifiers were designated AI-defined "Other" regions; this residual category was not trained as a cancer-associated fibroblast (CAF) classifier. Higher Other% was associated with shorter overall survival in univariable analyses, but the association was not significant after multivariable adjustment in TCGA. Multiomics and tissue-origin analyses linked Other% to stromal content. Scissor analysis of eight single-cell RNA-sequencing datasets identified Other-associated Scissor+ cells across epithelial and fibroblast compartments, while ACTA2 expression and extracellular matrix-remodeling pathways were enriched. In 48 MIBC specimens assessed by the hydrogel-embedded histoculture drug sensitivity test, Other% was higher in the gemcitabine-cisplatin-resistant group. Multiplex immunofluorescence further showed enrichment of alpha-smooth muscle actin signal in AI-defined Other regions. Erdafitinib showed ex vivo activity in a subset of specimens, but FGFR alteration status was unavailable and clinical efficacy was not evaluated. These findings support an association between AI-defined Other regions, CAF-related stromal features, and chemoresistance, while prospective validation is required before clinical use.
Journal IF-equivalent: 3.7 (OpenAlex 2-year mean citedness, value as of 2026-10-29, retrieved 2026-10-31; not the official Clarivate IF)Reference: Hu C, Wang F, Zhang Q, Xu H, Huang J, Xiong X, et al. Deep learning-based multiomics integration identifies AI-defined tissue regions associated with cancer-associated fibroblast enrichment and chemoresistance in muscle-invasive bladder cancer. Funct Integr Genomics. 2026 Oct 29;26(1):308. doi:10.1007/s10142-026-02086-6. PMID: 42955415.Checked: Abstract only