Abstract BACKGROUND Differentiating fibrotic from inflammatory strictures in Crohn’s disease can be challenging, but it is important for treatment decisions. Artificial intelligence (AI) models applied to magnetic resonance enterography (MRE) and computed tomography enterography (CTE) may improve noninvasive fibrosis assessment. We performed a systematic review and meta-analysis of histology-referenced studies evaluating the performance of these models for fibrosis. METHODS A systematic search was performed in PubMed and Embase from inception through August 2025 for human studies that evaluated AI or radiomics models using MRE or CTE to classify fibrosis in Crohn’s disease. Diagnostic performance was assessed by pooling area under the ROC curve (AUC) values across studies using a random-effects model. We also examined results by imaging modality and validation type, and performed sensitivity analyses to assess the influence of individual studies. RESULTS Four studies that included 698 bowel segments from 576 patients met our inclusion criteria (Table 1). Two studies utilized MRE with internal validation, while two used CTE with external validation. The pooled AUC for fibrosis classification was 0.85 (95% CI, 0.79-0.89; I2 = 84%) (Figure 1). Subgroup analysis showed AUCs of 0.87 (95% CI, 0.65-0.96) for MRE studies and 0.83 (95% CI, 0.80-0.85) for CTE studies. Sensitivity analysis excluding individual studies showed stable pooled values. Inflammation classification was assessed in only one study, which reported an AUC of 0.67. CONCLUSION AI models applied to MRE or CTE show moderate-to-high accuracy for detecting histologic fibrosis in Crohn’s disease strictures. Performance varies based on modality and validation type, with heterogeneity largely driven by study design and overlap between imaging approach and validation strategy. Larger multicenter studies with standardized protocols are needed before use in clinical practice.
Mussad et al. (Thu,) studied this question.