Multimodal deep learning for inflammatory bowel disease: a new frontier in cellular and molecular biomarker discovery to clinical translation

Inflammatory bowel disease (IBD), including Crohn’s disease (CD) and ulcerative colitis (UC), is a chronic, relapsing condition with heterogeneous clinical phenotypes and variable therapeutic outcomes. Deep learning (DL), combined with high-throughput sequencing and multi-omics, has advanced precision management by enabling integration of genomic, transcriptomic, microbiome, metabolomic, imaging, and clinical data. DL applications include molecular biomarker discovery, automated endoscopic and histopathological image analysis, patient stratification, disease monitoring, and prediction of therapeutic response, such as anti-TNF-α efficacy. Convolutional neural networks (CNNs) demonstrate exceptional performance in image interpretation and automated scoring. Challenges for clinical translation include limited multi-center datasets, inconsistent annotations, low interpretability, and privacy concerns. Addressing these issues through interpretable, efficient, and privacy-preserving DL frameworks, along with temporally resolved cross-institutional datasets, will facilitate real-time monitoring and personalized care, reshaping IBD diagnosis, treatment, and long-term management. Deep learning (DL) enables automated image analysis, multi-omics integration, and predictive modeling in inflammatory bowel disease (IBD). This review summarizes recent advances of DL in biomarker discovery, diagnosis, and therapeutic response prediction for IBD. Key challenges for clinical translation include data heterogeneity, model interpretability, and regulatory constraints. Strategies to accelerate clinical adoption include interpretable, efficient, privacy-preserving DL frameworks and cross-center data integration.