Semi-Automatic Assessment of Crohn's Disease Activity by Combined Analysis of Bowel Lesions and Creeping Fat

To develop and evaluate Crohn-BOOST, an open-source tool enabling semi-automatic segmentation of intestinal lesions and creeping fat on magnetic resonance enterography (MR Enterography), and to assess whether quantitative metrics derived from these segmentations relate to radiological disease activity scored with simplified MaRIA (sMaRIA) and Nancy indices. A fully annotated MR Enterography dataset was also curated to support future artificial intelligence research.

This retrospective single-center study included 102 patients with Crohn's disease (134 analyzable small-bowel lesions). Crohn-BOOST was used to delineate bowel lesions and creeping fat. Extracted metrics included lesion volume, wall thickness, lesion length, ADC, arterial enhancement, and creeping fat volume. Radiologic activity was assessed with sMaRIA and Nancy scores. Spearman correlations (95% CI by bootstrap), multivariable linear regression, and fivefold cross-validation were performed (p < 0.05).

Segmentation time was < 3 min per patient. Inter-reader reproducibility in a two-reader subset (n = 30) was high (Dice 0.85 for lesions; 0.87 for creeping fat). Lesion volume (sMaRIA r = 0.65; Nancy r = 0.61; p < 0.001) and ADC (sMaRIA r = -0.70) showed the strongest correlations with radiological activity. In multivariable models, these two metrics remained independent predictors, whereas creeping fat volume and enhancement metrics were not.

Crohn-BOOST enables fast, reproducible 3D quantification on routine MR Enterography and supports objective estimation of radiological disease activity. Lesion volume and ADC show strong associations with sMaRIA and Nancy scores, while creeping fat volume appears more related to chronic remodeling than acute inflammation.