Crohn’s disease: research progress in decoding pathogenic multi-network and precision management of artificial intelligence radiomics

Crohn’s disease (CD) is a chronic, relapsing inflammatory bowel disease characterized by transmural inflammation. Its clinical presentation and disease course are highly heterogeneous across individuals, and the global disease burden continues to rise. Although biomarkers such as fecal calprotectin and anti–Saccharomyces cerevisiae antibodies (ASCA), together with computed tomography enterography (CTE)/magnetic resonance enterography (MRE) and endoscopy, play central roles in diagnosis and longitudinal monitoring, important unmet needs remain. In particular, current approaches show limited reproducibility and insufficient phenotypic granularity for stratifying transmural inflammation, mesenteric involvement, and fibrostenotic disease, as well as for predicting therapeutic response and surgical risk. In this review, we adopt a multi-network pathogenic framework—encompassing genetic susceptibility, barrier dysfunction, microbial dysbiosis, immune-driven inflammation, fibrotic remodeling, and mesenteric inflammation with adipose remodeling—to delineate how these interconnected processes shape intestinal and mesenteric imaging phenotypes. We then focus on AI-enabled radiomics in CTE/MRE, summarizing key workflows for phenotype quantification, feature extraction, and model development, and highlighting its potential as an imaging biomarker across major clinical applications, including diagnosis and differential diagnosis, assessment of inflammatory activity, fibrosis stratification, prediction of treatment response, and surgical risk management. Importantly, rather than treating radiomics as a purely predictive “black box,” we organize current evidence within a mechanism-to-phenotype framework that links multi-network pathobiology and the histology/microenvironment to CTE/MRE imaging phenotypes and downstream radiomic signatures, thereby providing a biologically anchored basis for interpretation and model design. Finally, we discuss major challenges to clinical translation, including inter-center variability, differences in image acquisition and reconstruction, segmentation uncertainty, feature robustness, limited external validation, and clinical interpretability. We further outline a feasible roadmap for integrating radiomics with immunologic multi-omics to build a translatable evidence framework that supports precision management in CD.