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

Abstract

The integration of artificial intelligence (AI) into the histopathological assessment of inflammatory bowel disease (IBD) represents a fundamental paradigm shift toward a new era of computational pathology, promising unprecedented levels of objectivity, reproducibility, and clinical insight. Traditional semiquantitative histological scoring systems while clinically entrenched are intrinsically limited by significant inter-observer variability and subjective interpretation, complicating critical decisions in diagnosis, therapeutic monitoring, and the evaluation of histological remission, an increasingly paramount treatment goal. The advent of whole-slide imaging and sophisticated AI algorithms, particularly deep convolutional neural networks and vision transformers, offers a transformative solution. These technologies enable the automated, pixel-level analysis of mucosal architecture and inflammatory infiltrates, translating visual patterns into quantitative, continuous data. This comprehensive minireview systematically synthesizes the current state of AI applications in IBD histopathology, spanning automated crypt analysis, inflammatory cell quantification, and the prediction of established histological scores and clinical outcomes. It critically examines the potential of AI to operationalize and standardize the endpoint of histological remission, thereby strengthening clinical trials and precision medicine. Furthermore, the manuscript delves into the profound epistemological and practical challenges posed by this technological integration, including algorithmic bias, the “black box” dilemma, and the evolving role of the pathologist as an overseer of AI systems. Finally, it outlines a strategic roadmap for future translation, emphasizing the need for robust multicenter validation, standardized regulatory frameworks, seamless clinical workflow integration, and the cultivation of interdisciplinary expertise to fully realize the potential of AI in optimizing personalized, predictive care for patients with IBD.

Key Words: Artificial intelligence; Deep learning; Inflammatory bowel disease; Digital pathology; Histological remission; Precision medicine

Core Tip: Artificial intelligence is revolutionizing inflammatory bowel disease histopathology by using deep learning on whole-slide images to automate analysis. This addresses the subjectivity of current scoring systems, offering objective, quantitative data for diagnosis, monitoring, and defining histological remission. While promising for clinical trials and precision medicine, challenges like algorithmic bias and integration into pathology workflows must be overcome through validation, regulation, and interdisciplinary collaboration.

INTRODUCTION

Inflammatory bowel disease (IBD), primarily comprising Crohn’s disease (CD) and ulcerative colitis (UC), is a chronic, relapsing-remitting condition of the gastrointestinal tract, the pathogenesis of which involves a complex dysregulation of the immune response to environmental triggers in genetically susceptible individuals[1]. The clinical management of IBD has evolved significantly from symptom control toward a treat-to-target (T2T) strategy in which treatment decisions are guided by the achievement of objective, biomarker-driven endpoints to improve long-term outcomes[2]. Within this paradigm histopathological evaluation has ascended from a purely diagnostic tool to a central decision-making pillar, providing an indispensable microscopic window into disease activity and mucosal healing that complements clinical and endoscopic assessments[3]. This review focuses on IBD, as opposed to other gastrointestinal diseases, due to the unique and central role histology plays in its modern management. The chronic and relapsing nature of IBD necessitates frequent and precise histological assessment to guide therapy adjustments and define deep remission (histological healing), a key treatment target. This creates a high-stakes, repetitive clinical need for objective and reproducible measurement that is particularly amenable to AI augmentation. While AI applications are growing in other areas like Barrett’s esophagus or celiac disease, the imperative for standardized, quantitative histology in the IBD T2T paradigm presents a distinct and urgent opportunity for AI to deliver transformative clinical value.

Histology plays several non-negotiable roles. These include confirmation and differentiation of diagnosis (e.g., UC vs CD based on features like crypt architectural distortion or granulomas)[4]; assessment of activity and severity through grading acute inflammation, typically via quantification of neutrophilic infiltration[5]; and evaluation of therapeutic response to define histological remission-a state now recognized as a critical therapeutic target associated with improved long-term outcomes[6,7]. To standardize reporting several semiquantitative histological indices have been developed and validated, including the Geboes Score and Robarts Histopathology Index for UC and the Nancy Index and modified Global Histological Activity Score for CD[8-10]. These systems categorize mucosal findings into ordinal grades (e.g., 0-4) based on predefined criteria. However, this reliance on visual, subjective assessment by pathologists harbors substantial limitations that undermine its reliability. These include significant inter- and intra-observer variability, with kappa statistics often between 0.4-0.6[11,12]; reduction of a continuous biological spectrum into discrete categories, leading to loss of information; the labor-intensive nature of a thorough evaluation; and frequent overemphasis on acute inflammation at the expense of chronic features with potential prognostic significance[13]. These limitations create a critical gap between the clinical need for precise, reproducible histology, and the practical reality of its assessment. This gap represents the fundamental impetus for the integration of digital pathology and artificial intelligence (AI), which promises to augment human expertise with computational objectivity, transforming histology from an artisanal interpretation into a quantitative science.

METHODOLOGY

This minireview was conducted to synthesize the current evidence on AI applications in IBD histopathology. A systematic search of the literature was performed in PubMed/MEDLINE, IEEE Xplore, and Google Scholar databases for articles published between January 2015 and December 2024. Search terms included combinations of the following: (“artificial intelligence” OR “deep learning” OR “machine learning”) AND (“inflammatory bowel disease” OR “Crohn’s disease” OR “ulcerative colitis”) AND (“histopathology” OR “digital pathology” OR “whole slide imaging”). The initial search yielded 287 records. After removal of duplicates, titles and abstracts were screened for relevance. Full-text articles were assessed against the inclusion criteria of: (1) Primary application of an AI/machine learning method to histopathological images [whole-slide images (WSIs)] from human IBD samples; (2) Reporting of quantitative performance metrics; and (3) Publication in English language. Case reports, editorials, and studies focusing solely on endoscopy or radiology were excluded. Forty-two studies met the criteria and formed the evidence base for this synthesis. Given the narrative format, a formal meta-analysis was not conducted.

FOUNDATIONS: DIGITAL PATHOLOGY AS THE ENABLING SUBSTRATE

The digital transformation of pathology, initiated over two decades ago, is the essential precursor to AI integration. The creation of WSIs, high-resolution, digital facsimiles of entire glass slides, via dedicated scanners has decoupled the physical specimen from its microscopic evaluation[14]. This digitization offers foundational advantages such as facilitating remote consultation and the creation of large research databases, enabling the application of computer vision techniques for quantitative analysis, and potentially streamlining laboratory workflow. WSIs serve as the rich, structured data source upon which AI models are trained and deployed. The transition from optical microscopy to digital viewing is a necessary step toward a fully computational pathology workflow.

AI ARSENAL: TECHNIQUES AND APPLICATIONS IN IBD HISTOPATHOLOGY

At its core, AI in this context refers to computer systems, particularly deep learning (DL) models, that learn to recognize patterns in digital images. For instance, instead of a pathologist manually counting inflammatory cells, an AI model can be trained on thousands of annotated images to automatically identify, classify, and quantify different cell types across an entire biopsy slide, providing a precise numerical output. AI particularly DL has demonstrated remarkable success in image analysis tasks. In histopathology convolutional neural networks (CNNs) are the most widely used architecture. CNNs are designed to automatically learn hierarchical features from edges and textures to complex morphological structures directly from image pixels[15]. The applications of these techniques in IBD are rapidly expanding and can be categorized as follows (Table 1).

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.

Automated morphometric analysis: Beyond the human eye: This represents the most direct application in which AI performs quantitative measurements that are tedious or impossible for humans to do consistently at scale. For crypt architecture analysis, using U-Net or similar segmentation networks, AI can outline every crypt in a biopsy, calculating metrics like crypt density (number/mm²), crypt diameter, perimeter, and shape regularity. This allows for an objective assessment of architectural remodeling, a hallmark of chronicity in IBD that is poorly captured in standard activity scores[16,17]. For inflammatory cell quantification, AI models can be trained to detect and classify specific inflammatory cells after color deconvolution (separating hematoxylin and eosin signals). This enables precise counts of neutrophils (in lamina propria and epithelium), eosinophils, plasma cells, and lymphocytes per unit area or per crypt[18], reframing activity assessment from a subjective grade to a continuous variable.

Considering prediction of histological indices and clinical endpoints, AI aims to replicate and potentially improve upon expert human judgment. In automated histological scoring, several groups have developed end-to-end DL models that ingest a WSI and output a prediction for a standard score like the Geboes Score or Nancy index. Stidham et al[19], in the United States, demonstrated a CNN that could differentiate remission from active disease (Geboes ≥ 3B) with an area under the curve > 0.98. The most promising frontier is the prediction of clinical and endoscopic outcomes. AI can identify subtle morphological signatures indiscernible to humans that are prognostic. For example, Miyoshi et al[20], in Japan, developed a CNN that predicted the response to vedolizumab therapy in UC with significant accuracy by using baseline histology. Similarly, feature-based AI approaches have identified specific histologic structures, such as submucosal lymphoid aggregates, as predictors of postoperative recurrence in Crohn’s disease[21].

Operationalizing histological remission The definition of histological remission remains debated, often defined as the absence of neutrophils in the epithelium[22]. AI can bring clarity and standardization. By applying a universally consistent, quantitative threshold (e.g., ≤ 1 neutrophil per 10 high-power fields), AI can eliminate the ambiguity. Furthermore, AI can define more nuanced, multifeature remission signatures that incorporate architectural and chronic inflammatory components, potentially identifying a more robust complete healing phenotype associated with the best outcomes[23].

NAVIGATING THE CHASM: EPISTEMOLOGICAL, TECHNICAL, AND PRACTICAL CHALLENGES

The integration of AI is not merely a technical upgrade but a philosophical shift in diagnostic epistemology, accompanied by significant hurdles. Epistemological and ethical questions remain. Key challenges include explainability and trust, as many high-performing DL models are “black boxes”, making it difficult to understand their predictions, which in turn hinders clinical trust[24,25]. Algorithmic bias and generalizability are also major concerns; a model trained on data from a single center or specific demographic may perform poorly or inequitably elsewhere[26-30]. The current evidence base, as seen in Table 2[31-34], is largely from research centers in North America, Europe, and East Asia, highlighting the need for more geographically diverse training data to ensure global applicability. Furthermore, the role of the pathologist will evolve toward that of a curator and overseer of AI systems[27].

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.

Technical and logistical hurdles need to be acknowledged. These include the need for robust computational infrastructure for storing and processing WSIs, the cruciality of standardizing tissue staining and scanning protocols to ensure model robustness, and the inherent challenge of navigating complex regulatory pathways for AI-based software as a medical device[28].

DISCUSSION: FUTURE HORIZONS AND IMPLEMENTATION BARRIERS

The future trajectory of AI in IBD histopathology is promising but complex. Firstly, as AI technology evolves, we can envision it moving beyond replicating human scores to discovering novel, prognostically significant histopathological patterns (“histophenotypes”) invisible to the human eye. In the near future, integrated AI systems could provide real-time, quantitative histological data during endoscopy, informing therapeutic decisions instantly, and continuously monitor disease activity through longitudinal biopsy analysis to predict flares or treatment failure. Secondly, ethical considerations are paramount. Beyond algorithmic bias, these include patient data privacy in large WSI repositories, informed consent for the use of biopsies in AI development, and establishing clear medico-legal accountability for AI-assisted diagnoses. Developing transparent, auditable AI and inclusive ethical frameworks is essential. Thirdly, the economic cost is substantial but must be evaluated holistically. Significant initial investment is required for digital scanners, data storage, AI software, and information technology (IT) support. However, potential value may be realized through increased pathologist efficiency (freeing them for complex cases), reduced diagnostic variability (leading to fewer unnecessary treatments or procedures), and improved patient outcomes from precision management (which could lower long-term healthcare costs). Comprehensive cost-effectiveness studies are needed. Finally, significant institutional and cultural barriers exist. Resistance to change within established pathology workflows, lack of specific reimbursement models for AI-assisted diagnosis, interoperability challenges between AI platforms and hospital IT systems (Laboratory Information Systems, Electronic Health Records), and a shortage of professionals with hybrid expertise in pathology and data science are key hurdles. Overcoming these requires proactive leadership, policy innovation, and the integration of digital pathology and AI fundamentals into medical and pathology training curricula.

FUTURE PERSPECTIVES AND ROADMAP FOR INTEGRATION

For AI to transition from promising research to routine clinical practice, a coordinated, multistakeholder effort is required.

Phase 1: Robust validation and benchmarking

Establishment of large, public, expertly annotated WSI datasets for IBD (similar to The Cancer Genome Atlas for oncology) to serve as universal benchmarks is needed. Prospective, multicenter clinical trials must evaluate AI tools not just for technical accuracy but for clinical utility.

Phase 2: Standardization and regulation

Development of consensus guidelines for AI development in pathology (e.g., extending the MINIMAR checklist)[29] is needed. Regulatory bodies need to create clear, adaptive pathways for Software as a Medical Device that balance innovation with patient safety.

Phase 3: Workflow integration and education

AI tools must be embedded seamlessly into the digital pathology information system and electronic health record. Concurrently, medical education curricula must evolve to train the next generation of gastroenterologists and pathologists in digital and computational pathology principles.

Phase 4: Advanced discovery and personalization

Looking ahead, AI will enable the discovery of entirely novel, quantitative histophenotypes. Integrating histology-based AI with other multi-omics data (genomics, transcriptomics from the same biopsy) could unlock powerful, multidimensional biomarkers for truly personalized therapy selection and disease subtyping[30].

CONCLUSION

AI is poised to revolutionize the histopathological assessment of IBD. By delivering quantitative, reproducible, and objective analyses, it directly addresses the core limitations of conventional microscopy. Its applications range from automating routine scoring to predicting therapeutic response, thereby strengthening the pillars of the T2T strategy and bringing the goal of precision medicine closer to reality. However, this journey is fraught with epistemological, technical, ethical, and practical challenges that must be thoughtfully addressed. The successful integration of AI will depend not on the blind pursuit of automation but on a collaborative, human-centric model in which computational power augments clinical expertise. By navigating this path carefully, the gastroenterology and pathology communities can harness AI to standardize care, deepen biological understanding, and ultimately improve the long-term prognosis and quality of life for patients living with IBD.