Artificial intelligence in histopathology of inflammatory bowel disease: Toward objective and reproducible assessment of disease activity

Table 1 Core artificial techniques and their applications in inflammatory bowel disease histopathology

AI technique	Description	Primary application in IBD	Key advantage
Semantic segmentation	Classifies each pixel in an image into a predefined class (e.g., crypt epithelium, lamina propria, lumen)	Crypt segmentation and architectural analysis; Quantifies crypt density, distortion, branching, and atrophy	Provides a comprehensive, structural map of the mucosa; Enables precise measurement of architectural parameters
Instance segmentation	Identifies and delineates each individual object instance (e.g., each separate crypt, each inflammatory cell)	Individual crypt isolation and inflammatory cell detection/quantification	Allows for per-object analysis (size, shape of each crypt) and precise cell counting (neutrophils, eosinophils)
Object detection	Identifies and locates objects within an image using bounding boxes	Rapid identification of regions of interest, such as areas with severe activity or ulceration	Efficiently guides pathologist attention or focuses deeper analysis on most relevant areas
Whole-slide classification	Assigns a single label or score to an entire WSI	Direct prediction of global histological scores (e.g., Geboes ≥ 3.0, Nancy ≥ 2) or remission status	Automates scoring workflow, reduces time-to-diagnosis, and standardizes output

AI: Artificial intelligence; IBD: Inflammatory bowel disease; WSI: Whole-slide imaging.

Table 2 Summary of key studies on artificial intelligence in inflammatory bowel disease histopathology

Ref.	AI task	Cohort (disease)	Key finding	Implication
Rymarczyk et al[31], 2024	Crypt segmentation using U-Net	385 WSIs (UC and CD)	Quantified crypt distortion correlated with endoscopic severity and predicted clinical outcomes	AI provides objective architectural metrics of chronic damage
Rymarczyk et al[31], 2024	WSI classification for Geboes score	913 WSIs (UC)	CNN achieved AUC > 0.98 for discriminating active disease (Geboes ≥ 3B)	High accuracy in automating a complex histological score
Minea et al[32], 2025	WSI classification for therapy response	161 patients (UC)	Baseline histology-based CNN predicted vedolizumab response (AUC 0.79)	Histology contains prognostic signals for biologic therapy outcomes
Villanacci et al[33], 2023[33]	Feature-based model for relapse risk	88 patients (CD)	AI-quantified density of submucosal lymphoid aggregates predicted post-surgical relapse	Identified a novel histologic prognostic biomarker
Rubin et al[34], 2025	Multicenter validation of a Nancy Index predictor	583 WSIs from five centers (UC)	Model generalized well across centers (weighted kappa 0.70 with experts)	Demonstrates potential for cross-institutional standardization

AI: Artificial intelligence; AUC: Area under the curve; CD: Crohn’s disease; CNN: Convolutional neural network; UC: Ulcerative colitis; WSI: Whole-slide imaging.