Rediscovering histology – the application of artificial intelligence in inflammatory bowel disease histologic assessment

Abstract

Integrating artificial intelligence (AI) into histologic disease assessment is transforming the management of inflammatory bowel disease (IBD). AI-aided histology enables precise, objective evaluations of disease activity by analysing whole-slide images, facilitating accurate predictions of histologic remission (HR) in ulcerative colitis and Crohn’s disease. Additionally, AI shows promise in predicting adverse outcomes and therapeutic responses, making it a promising tool for clinical practice and clinical trials. By leveraging advanced algorithms, AI enhances diagnostic accuracy, reduces assessment variability and streamlines histological workflows in clinical settings. In clinical trials, AI aids in assessing histological endpoints, enabling real-time analysis, standardising evaluations and supporting adaptive trial designs. Recent advancements are further refining AI-aided digital pathology in IBD. New developments in multimodal AI models integrating clinical, endoscopic, histologic and molecular data pave the way for a comprehensive approach to precision medicine in IBD. Automated assessment of intestinal barrier healing – a deeper level of healing beyond endoscopic and HR – shows promise for improved outcome prediction and patient management. Preliminary evidence also suggests that AI applied to colitis-associated neoplasia can aid in the detection, characterisation and molecular profiling of lesions, holding potential for enhanced dysplasia management and organ-sparing approaches. Although challenges remain in standardisation, validation through randomised controlled trials and ethical considerations. AI is poised to revolutionise IBD management by advancing towards a more personalised and efficient care model, while the path to full clinical implementation may be lengthy. However, the transformative impact of AI on IBD care is already shining through.

Plain language summary

Enhancing IBD tissue analysis through artificial intelligence

Artificial intelligence (AI) is significantly advancing in treating inflammatory bowel disease (IBD). By analysing tissue samples, AI can accurately assess disease activity and predict outcomes, helping physicians personalise treatment plans and improve patient care in clinical practice. In clinical research, AI enhances trials by quickly analysing tissue samples and potentially identifying patients likely to respond to treatments. As AI develops, its potential to integrate various data sources promises a future where IBD treatment is more precise, efficient, and personalised.

Infographic: AI-aided pathology in IBD. This infographic provides a schematic overview of the capabilities of AI applied to histology in IBD. It highlights AI’s potential for disease assessment, predictions of outcomes and responses to therapy, integration and harmonisation of multimodal data, automated evaluation of barrier healing and detection and characterisation of IBD-related lesions. In the lower section, the infographic summarises the transformative impact of AI on clinical trials and routine clinical practice, showcasing its potential to revolutionise IBD management.

Source: Created with ‘Biorender.com’.

AI, artificial intelligence; IBD, inflammatory bowel disease.

Keywords

clinical trial deep-learning endo-histo-OMIC histologic remission machine learning precision medicine response to therapy

Introduction

Inflammatory bowel disease (IBD), which includes ulcerative colitis (UC) and Crohn’s disease (CD), is a major cause of disability worldwide and is becoming increasingly prevalent.¹ Clinical and endoscopic disease activity is closely linked with adverse outcomes in IBD patients, establishing clinical remission and endoscopic healing as formal treatment targets in both UC and CD.² However, the evolving concept of deep remission has expanded IBD treatment goals beyond endoscopic healing to include histologic, transmural and barrier healing.³ Recent studies have underscored the importance of deeper remission in IBD management^4–7 with a renewed focus on histology as a fundamental aspect of disease control. Achieving histologic remission (HR) in IBD has been associated with reduced corticosteroid use, lower hospitalisation and surgery rates and a decreased risk of IBD-related neoplasia. The emerging concept of ‘endo-histo mucosal healing’ is gaining traction in UC clinical trials^8,9 and future studies aim to define more comprehensive endpoints, such as transmural and barrier healing, advancing the pursuit of ‘deep healing’ in IBD.

Although several histological scoring systems have been developed for UC and CD, they remain limited in clinical use, with a universally accepted definition of HR still an unmet need.¹⁰ Currently, only three scores for histological inflammation in UC are fully validated, that is, the Robarts Histopathology Index (RHI), Nancy Histological Index (NHI) and Paddington International Virtual Chromoendoscopy Score Histologic Remission Index (PHRI).^11–13 The Geboes Score (GS) is widely used in clinical trials despite lacking formal validation.¹⁴ No scoring systems for CD have been fully validated, but the Global Histological Activity Score (GHAS) remains commonly used.¹⁴ The latest European Crohn’s and Colitis Organisation (ECCO) guidelines recommend that histological scoring systems developed for UC, including RHI, NHI and GS, may also be applied to intestinal biopsies in CD patients.⁶ These scores typically evaluate inflammatory cell infiltrates, but determining the exact prognostic value of each histologic feature is challenging. While the role of eosinophils in UC and CD has been downplayed due to a lack of consensus on their prognostic significance,^15,16 neutrophils are a focal point, with the absence of neutrophilic infiltration now uniformly recognised as a key indicator of HR.^7,17 The recently developed, simplified PHRI score for UC, which relies solely on the presence or absence of neutrophils in the lamina propria or epithelium, has shown a strong correlation with endoscopic activity, various histologic scores and long-term clinical outcomes.¹³

Despite these advancements, challenges remain, including limited real-world adoption of histological scoring systems, high inter- and intra-observer variability and the need for specialised expertise. A recent global survey revealed that most histological reports in clinical practice are descriptive and lack standardisation.¹⁸ While the accuracy of these scoring systems is highest when assessed by expert pathologists, the process is time-consuming and subject to variability.¹⁹ Moreover, the inherent subjectivity in histologic interpretation can compromise result precision and reproducibility, posing a significant challenge in IBD management. In this context, artificial intelligence (AI) has demonstrated substantial potential to revolutionise IBD management.²⁰ AI offers the opportunity to address these challenges by enabling rapid, accurate and standardised assessments of histologic disease activity.

This narrative review explores the potential application of AI in histologic assessment for IBD in both clinical practice and trials, focusing on its transformative potential to enhance outcome prediction and integrate endoscopy and histology data with multi-omics datasets. This integration sets the stage for a more personalised approach to IBD management.

AI-aided histology for disease assessment

A significant application of AI in IBD is assessing histological disease activity, where AI focuses on specific histological features to provide standardised, objective and efficient evaluations. This approach enhances disease management by reducing variability and increasing accuracy in histologic interpretation, ultimately improving patient outcomes (Figure 1). Table 1 summarises key studies investigating AI models applied to histology in IBD.

Figure 1.

AI-enabled digital pathology in IBD. This figure illustrates the current applications of AI in histological assessment for IBD. AI has demonstrated significant accuracy and efficiency in assessing disease activity, enabling standardised, rapid tissue segmentation and precise cell detection and characterisation. These capabilities position AI as a powerful tool in IBD, facilitating the evaluation of inflammation and healing, as well as enhancing tissue characterisation. Ultimately, AI contributes to improved prediction of patient outcomes and responsiveness to therapy. In the upper right panel, AI-enabled detection of disease activity is depicted (inflammation in red). The central panel illustrates tissue segmentation, distinguishing superficial epithelium (light blue), lamina propria (dark blue) and crypts (green, with crypt lumen in yellow). In the lower right panel, automated cell detection and characterisation are shown, identifying epithelial cells (green), neutrophils (orange), lymphocytes (red), plasma cells (light blue) and stromal cells (yellow).

Table 1.

Summary of key studies exploring AI models applied to histology in IBD.

Study	Design	Population	Primary outcomes	ICC pathologists/AI tool	Precision	Sensitivity	Specificity	Accuracy	AUC
Matalka et al. (2013)²¹	Retrospective single centre	118 biopsies	Assess mucosal damage and architectural distortion in IBD		98.3%
Vande Casteele et al. (2022)²²	Retrospective single centre	88 biopsies from UC patients	DL to identify eosinophils	80.5%–91.7%		86.4%
Gui et al. (2022)¹³	Prospective multicentre	614 biopsies from 307 UC patients	AI system to predict histological remission (PHRI 0 vs PHRI >0)			78%	91.7%	86%
Kiyokawa et al. (2022)²³	Retrospective single centre	68 CD patients	CNN to predict CD postoperative recurrence		96.4%	96.5%		96.9%	0.995
Del Amor et al. (2022)²⁴	Prospective multicentre	230 colorectal biopsies from UC patients	DL to determine the presence of UC activity based on neutrophils detection		90%	96%	83%	91%	0.898
Ohara et al. (2022)²⁵	Retrospective single centre	WSI from 114 UC patients in clinical and endoscopic remission	DL model to detect goblet cell mucus area and predict prognosis			93%; 83%	96%; 98%	95; 97%
Najdawi et al. (2023)²⁶	Retrospective single centre	512 WSI from 334 UC patients (training) and 308 WSI from 243 UC patients (testing)	CAD system to predict NHI and histological remission (NHI ⩽1 vs 2–4)					97%
Rymarczyk et al. (2023)²⁷	Retrospective multicentre	2935 biopsies from 302 CD patients and 3496 biopsies from 887 UC patients	AI to predict GHAS in CD; AI to predict GS for UC					80%–89%; 65%–85%
Iacucci et al. (2023)²⁸	Prospective multicentre	535 biopsies from 273 UC patients	CAD system to distinguish histological remission/activity for RHI, NHI and PHRI			94%; 89%; 89%	76%; 79%; 86%	80%, 81%, 90%	0.85; 0.86; 0.90
Liu et al. (2024)²⁹	Retrospective multicentre	292 UC patients (training) and 113 UC patients (testing)	AI model to predict CSFR in paediatric UC patients on mesalamine therapy			84%; 78%	94%; 91%	90%; 85%	0.89; 0.85
Wang et al. (2024)³⁰	Retrospective multicentre	278 WSI from 205 CD patients	DL system to predict CD postoperative recurrence			75%–89.1%	88%–96.8%	84%–93.5%	0.87–0.98
Peyrin-Biroulet et al. (2024)³¹	Retrospective single centre	200 histopathology slides from UC patients	AI system to measure histological disease activity based on NHI	87.2%
Furlanello et al. (2024)³²	Prospective single centre	356 biopsies from 52 patients (24 CD, 19 UC, 9 HC)	DL to identify basal plasmacytosis to support the diagnosis of IBD					90%
Ohara et al. (2024)³³	Prospective single centre	WSI from 96 UC patients	AI system to quantify and localise neutrophils and predict clinical relapse		77%	81%
Reigle et al. (2024)³⁴	Retrospective multicentre	530 crops from 143 rectal mucosal biopsies from paediatric UC patients	DL to identify and count eosinophilic cells		77%	96%		86%
Rubin et al. (2024)³⁵	Prospective multicentre	630 WSI (training) and 162 WSI (testing) from UC patients	AI tool to measure histological disease activity based on NHI		96%

AI, artificial intelligence; AUC, area under the curve; CAD, computer-aided diagnosis; CD, Crohn’s disease; CNN, convolutional neural network; CSFR, corticosteroid-free remission; DL, deep learning; GHAS, global histology activity score; GS, Geboes score; IBD, inflammatory bowel disease; ICC, intraclass correlation coefficient; NHI, Nancy histological index; PHRI, PICaSSO Histologic Remission Index; PICaSSO, Paddington International Virtual Chromoendoscopy Score; RHI, Robarts histopathology index; UC, ulcerative colitis; WSI, whole slide image.

Disease assessment in UC

One of the earliest AI studies applied to UC, conducted by Vande Casteele et al.²² developed deep learning (DL) algorithm to analyse whole-slide images (WSIs) of colon biopsies. The study identified links between eosinophil density and histological and clinical characteristics in patients with active UC. Gastrointestinal pathologists annotated eosinophil counts from digitised colon biopsies of 88 UC patients, which were used to train the model. The algorithm achieved a sensitivity of 86.4%, an accuracy of 91.8% and an F1 score (harmonic mean of precision and recall) of 0.89. During validation, eosinophil counts from 20 areas not used in training showed near-perfect agreement with manual counts, revealing correlations between eosinophil density, histological activity and disease extent.

Gui et al.¹³ subsequently developed the PHRI, a simple metric for assessing inflammation in UC, and created an AI diagnostic system to quantify it. Trained on 138 UC biopsy slides with varying inflammation levels, the system used a convolutional neural network (CNN) to detect neutrophils, calculate PHRI and classify biopsies into HR or non-remission. During testing and validation, the AI system achieved accuracies of 0.89 and 0.87 for neutrophil detection and 0.86 and 0.93 for disease activity classification, demonstrating a robust ability to predict HR. Notably, the same group enhanced the model using a larger dataset of 535 UC biopsies to predict disease flare, achieving over 80% accuracy in predicting endoscopic inflammation, with strong agreement between AI and human pathologists.²⁸

Similarly, Najdawi et al.²⁶ developed CNN models to quantify histological features in UC from WSI images, assessing disease activity through tissue segmentation and cell classification. These models identified cell types, including neutrophils and plasma cells, generating human-interpretable features that correlated strongly with disease severity and pathologist-assigned histological scores. Using these features, a random forest classifier accurately predicted histological index scores and remission status based on neutrophil activity, achieving an impressive 0.97 accuracy for predicting HR.

More recently, Ohara et al.³³ developed an AI system using semantic segmentation and object detection models to identify neutrophils in haematoxylin and eosin-stained WSIs. This system detects the presence and location of neutrophils within the epithelium and lamina propria and predicts the NHI and PHRI components. The model demonstrated strong performance metrics, with precision, recall and F1 score values of 0.77, 0.81 and 0.79, respectively. Additionally, histological scores predicted by the AI system positively correlated with pathologists’ diagnoses (Spearman’s ρ = 0.68–0.80; p < 0.05).

Lastly, Peyrin-Biroulet et al.³¹ applied an automated image analysis approach combined with machine learning algorithms to assess histological disease activity based on the NHI in 200 histological images from UC patients. The performance of the AI system was compared to that of four independent histopathologists. Despite limitations in the annotated training image dataset, the authors reported strong correlations among histopathologists and between histopathologists and the AI system.

AI is showing significant promise in advancing UC histological assessment by providing accurate, efficient and standardised evaluations. Our recent meta-analysis and meta-regression evaluated AI performance in assessing HR compared to pathologists.³⁶ AI models showed strong performance, with no significant difference with pathologists for specificity, observed agreement and F1 score, despite pathologists achieving the highest sensitivity and negative predictive value. This underscores AI’s potential to enhance clinical decision-making and disease management. However, further research is required to refine AI models and validate their utility in larger, real-world patient cohorts.

Disease assessment in CD

In CD, most studies have centred on developing DL systems aimed at a differential diagnosis, detecting pathological signs and predicting therapeutic responses, with fewer efforts focused on analysing specific histological features.

Rymarczyk et al.²⁷ used scanned histology slides from clinical trials in CD and UC to train AI models to predict the GHAS for CD and the GS for UC. The model, which used multiple instance learning and an attention mechanism (SA-AbMILP), outperformed others, achieving accuracies between 65% and 89%. It effectively differentiated microscopic disease activity by identifying critical histological features, with accuracies of 87%–94% in the colon for both diseases and 76%–83% in the CD ileum.

Furlanello et al.³² introduced an AI-based scoring system to assist in diagnosing IBD by semi-automatically quantifying basal plasmacytosis. Their model, based on the StarDist algorithm, analysed 4981 annotated images from a public dataset and validated its performance with 356 biopsies from CD, UC and healthy control patients. The DL model closely matched human assessments of basal plasmacytosis, a key diagnostic feature distinguishing IBD from non-IBD colitis, demonstrating AI’s diagnostic potential in IBD.

Recent advancements have explored AI-aided histology for post-operative CD assessment. Kiyokawa et al.²³ conducted the first DL analysis of histological images to predict CD recurrence post-surgery. Their model, which analysed 550 WSIs, identified adipocyte shrinkage and mast cell infiltration as primary predictors of recurrence, achieving high accuracy (AUROC 0.995).

More recently, Wang et al.³⁰ developed and validated a DL system to predict post-operative recurrence by automatically identifying features in the muscular layer and myenteric plexus. Their multitask model, incorporating clinical data, accurately assessed myenteric plexitis severity and CD recurrence risk, providing insights into post-operative recurrence pathogenesis and potential strategies for improving long-term outcomes.

AI is emerging as a valuable tool for assessing histology in CD, with promising applications in diagnosis, disease activity prediction and postoperative assessment. While more evidence is needed compared to its application in UC, these models pave the way for streamlined disease monitoring and support personalised treatment strategies in CD.

AI-aided histology for outcome prediction in IBD

AI-aided histology in IBD has shown strong predictive capabilities for adverse outcomes and response to therapy, especially in UC. A computer-aided diagnosis system developed by Iacucci et al.²⁸ demonstrated high sensitivity and specificity in predicting the 12-month flare risk among UC patients. Utilising the PHRI to distinguish between patients with histological activity and those in HR, the model achieved a hazard ratio for disease flare-up (hazard ratio 4.64), comparable to that of human pathologists (hazard ratio 3.56).

Similarly, Ohara et al.³³ used their AI system to score WSIs using NHI and PHRI to predict clinical relapse risk in UC patients. Higher AI-generated scores corresponded to hazard ratios of 3.2–5.0, indicating an elevated relapse risk. Previously, the same team demonstrated that a DL model quantifying goblet cell mucus area could effectively predict clinical relapse in UC patients in clinical and endoscopic remission.²⁵ Patients identified by the model as having goblet cell depletion had a significantly higher relapse rate compared to those without depletion (45% (10/22) vs 6.5% (6/92); p < 0.01).

In addition, Liu et al.²⁹ conducted a promising study exploring AI-aided histology for predicting therapeutic response. Their machine learning model, using 18 histologic features in paediatric UC patients, accurately predicted steroid-free remission on mesalamine therapy, underscoring AI’s potential for tailoring treatment strategies in IBD.

While applications of AI in UC continue to expand, AI-based histology for outcome prediction in CD remains an unmet need. Nevertheless, the potential of AI to predict outcomes could transform clinical management and personalising treatment strategies in IBD.

Application of AI-aided histology in clinical trials

Endoscopic remission and HR are increasingly recognised as key endpoints in clinical trials, reflecting deeper disease control than clinical and endoscopic remission alone.^37,38 Accurate and objective evaluation of histological activity through standardised and validated scoring systems is essential to optimising treatment allocation and assessing therapeutic efficacy.

As previously noted, histological scoring systems such as the GS for UC and the GHAS for CD are commonly used in clinical trials to assess histological endpoints. However, these systems lack full validation, and their inconsistent application can lead to variability in histological assessment across pathologists.

To address these challenges, clinical trials often rely on central reading by expert pathologists to achieve more consistent and standardised evaluations.³⁹ However, even among experienced professionals, interobserver variability can persist,¹⁹ potentially introducing inconsistencies in trial results and affecting the interpretation of findings. AI-based systems offer a solution by providing accurate, objective and reproducible histological evaluations, significantly reducing variability and enhancing the reliability of histological endpoints.²⁰

AI models can match or even exceed pathologists’ performance in grading histological activity, with enhanced sensitivity to subtle changes that the human eye might overlook. For instance, the DL model developed by Rymarczyk et al. for automated histological assessment in IBD used phase II and III clinical trial images and achieved moderate to substantial agreement with central readings. This model reliably identified key histological features, performing comparably to independent pathologists.²⁷

The real-time capabilities of AI are particularly advantageous in adaptive trial designs, where early indicators of response can dynamically inform treatment strategies. Traditional histopathological analysis is often completed after collecting samples, resulting in considerable delays. By contrast, AI can analyse biopsy samples in real-time as they are collected, offering immediate insights to expedite decision-making.⁴⁰ Furthermore, the ability of AI to rapidly process large volumes of data makes it especially suited for large-scale trials, improving efficiency and reducing the time and costs associated with manual evaluation.

Notably, Iacucci et al.⁴¹ recently developed an innovative AI model to assess early response to therapy using WSIs from the phase II Mirikizumab (anti-IL-23 monoclonal antibody) clinical trial. This deep neural network model demonstrated an impressive ability to predict response at 12 and 52 weeks, based on histological remission and improvement as measured by the GS. This study paves the way for AI-driven decision-making in clinical trials, assisting in optimising drug and dosage choices and providing objective support for personalised management in UC patients.

Additionally, AI-driven analysis can integrate with other data sources, such as clinical, laboratory and endoscopic findings, to create a more comprehensive disease evaluation. This integrative approach can potentially identify new histological markers that could serve as secondary endpoints in clinical trials.²⁰ Recently, Harun et al.⁴² developed a machine learning model using data from the phase III Etrolizumab (anti-β7 integrin monoclonal antibody) clinical trial. By leveraging patient-level data on demographics, physiology, disease history, clinical questionnaires, histology, serum biomarkers and Etrolizumab exposure, they identified key predictors of remission. This preliminary evidence underscores the potential of AI-enabled stratification in the design of future clinical trials.

In conclusion, AI-aided histology is promising to enhance histological assessments’ precision, speed and efficiency in clinical trials. By improving the consistency of evaluations, integrating diverse data sources and enabling real-time analysis, AI can revolutionise clinical trial design and the development of personalised treatment strategies in IBD.

Application of AI-aided digital pathology in clinical practice

AI-assisted histological evaluation can transform everyday clinical practice for IBD monitoring. Accurate and objective assessment of histological activity is essential for effective disease management, as HR is associated with improved long-term outcomes compared to clinical and endoscopic remission alone.^43,44 Current ECCO guidelines recommend using standardised scoring systems, such as the RHI, NHI and PHRI, to evaluate histological activity.^6,7 However, these systems are not widely adopted in clinical practice due to their complexity, the limited availability of specialised training and the time required for their application.¹⁸ Additionally, manual scoring by pathologists can be subjective and inconsistent, which may impact clinical decision-making.

Integrating AI technologies into histopathological workflows offers a promising solution to these limitations, enhancing diagnostic accuracy, reducing interobserver variability, accelerating routine evaluations and improving patient outcomes.⁴⁵

Several studies have shown that AI algorithms can analyse and grade histological activity with accuracy comparable to that of expert pathologists, making them valuable additions to clinical workflows, especially where rapid diagnostics are essential.⁴⁶ Beyond standardisation, AI can also enhance human expertise by detecting subtle histological changes that may be overlooked by non-expert pathologists, particularly in non-referral centres. Early detection of these changes enables clinicians to make timely treatment adjustments, optimising therapeutic strategies based on a more precise understanding of histological disease activity.⁴⁷

AI systems can also significantly reduce the time required for histological evaluations, making them ideal for high-volume clinical settings. Traditional histopathological analysis is labour-intensive and time-consuming, especially when handling large numbers of biopsy samples. By automating the detection and quantification of key histological features, AI tools can expedite turnaround times, allowing clinicians to make faster, data-driven decisions.

Furthermore, integrating AI into histological analysis paves the way for personalised medicine. AI tools can analyse histological images and combine findings with clinical, laboratory and imaging data to provide a comprehensive disease assessment, guiding tailored treatment strategies.²⁰

Overall, AI-aided histology is positioned to revolutionise clinical practice in IBD by enhancing diagnostic precision, increasing workflow efficiency and supporting personalised treatment. As these technologies continue to advance, they promise to significantly improve the standard of care and long-term outcomes for IBD patients.

Going beyond AI-aided histology towards precision medicine

The application of AI in IBD pathology is advancing beyond the simple assessment of histological activity, reaching exciting new deep levels of precision medicine.

The innovative ‘Endo-Histo-OMICs’ approach, recently proposed by Iacucci et al.²⁰ represents an encouraging development in IBD management. The AI-enabled integration and harmonisation of multimodal data, encompassing endoscopy, histology and OMICs, are promising to revolutionise IBD care. It offers multiple potential applications in clinical practice, including patient profiling, early diagnosis, biomarker discovery, guiding clinical outcome prediction and personalised treatment. An exploratory study involving 29 IBD patients (15 CD and 14 UC) has already demonstrated the benefit of integrating advanced imaging, such as probe confocal laser endomicroscopy (pCLE), with transcriptomics data to predict response to biological therapy.⁴⁸ Specifically, endomicroscopy mucosal and microvascular changes detected by AI-aided pCLE – including vessel tortuosity, crypt morphology and fluorescein leakage – integrated with transcriptomic-identified genes, including ACTN1, CXCL6, LAMA4, EMILIN1, CRIP2, CXCL13 and MAPKAPK2, predictive of response to biological agents. Moreover, this multimodal approach might also embody vast potential in future clinical trials, guiding the assessment of deep remission, selecting the optimal treatment for individual patients, optimising drug selection, dosage and duration, reducing adverse events and expediting drug development. The introduction of foundation models, capable of processing large multimodal datasets with minimal task-specific adjustments, and generative AI, which learns underlying patterns in data to generate novel outputs, presents unique opportunities for the future.^49–51 These models can further accelerate and standardise the ‘Endo-Histo-OMICs’ approach, bringing it closer to routine clinical use in trials and practice.

AI is advancing towards a deeper level of assessment in intestinal barrier healing. Recent studies have shown that barrier healing is a better predictor of patient outcomes than endoscopy and histology,^52,53 paving the way for deeper levels of healing and more personalised management of IBD patients. Several new tools have emerged for assessing the intestinal barrier at the cellular and molecular level, including advanced endoscopic and laboratory techniques.⁵⁴ However, their broader application is limited by reduced availability, high costs and complexity, requiring specialised expertise. If standardised, automated and integrated by AI, these tools can refine intestinal barrier assessment, improving the accuracy of diagnosis, prognosis and disease management. Recently, multispectral intestinal barrier imaging, guided by ultra-high magnification endocytoscopy, exhibited promise in identifying barrier healing and predicting major adverse outcomes over 12 months in UC patients.⁵⁵ This approach may enhance the treat-to-target strategy in UC. Moreover, machine learning models offer the opportunity to identify barrier-protective therapies and predict candidate agents for clinical trials. AI has demonstrated the ability to predict epithelial barrier-related genes, such as PRKAB1, the β1 subunit of the metabolic master regulator, AMPK, which might represent a novel target for gut barrier-protective therapies.⁵⁶

Finally, it is worth highlighting the development of AI models capable of accurately and rapidly detecting and characterising IBD-related lesions. Recently, Abdelrahim et al.⁵⁷ developed and validated a novel AI model that demonstrated remarkable performance in detecting and characterising endoscopic lesions from 30 IBD patients. Although evidence on AI-enabled digital pathology for IBD-associated lesions is still lacking, AI exhibits the capacity, as a valuable tool, for accurately analysing histological slides and aiding pathologists in identifying and characterising subtle lesions.⁵⁸ Moreover, these models can potentially integrate histology with other modalities – such as endoscopic and molecular data – leading to deeper, more precise lesion characterisation and enabling personalised, effective treatment strategies and an organ-sparing approach. Intriguingly, AI combined with digital pathology has been applied to predict p53 immunohistochemical staining results – a key marker of IBD-related dysplasia⁵⁹ – from UC-associated neoplasia.⁶⁰ The model achieved impressive performance metrics in predicting p53, offering a cost-effective, time-efficient and detailed molecular characterisation tool that could guide the management of colorectal lesions.

In conclusion, the integration of AI in IBD is ushering in a new era of precision medicine. By harmonising multimodal data, AI transforms patient care through early diagnosis, accurate patient profiling and therapeutic response prediction. The ability to assess deeper levels of healing, such as barrier healing, precise lesion detection and characterisation, will improve treatment strategies and potentially enable organ-sparing approaches.

Integrating AI in IBD pathology: challenges and opportunities

While AI applied to histology in IBD has demonstrated clear advantages, including rapid and objective histological assessment, accurate prediction of outcomes and standardised analysis of large datasets for a comprehensive approach to IBD management, several challenges remain. The implementation of AI into clinical practice requires addressing these hurdles.²⁰

Data quality used to train and test AI is crucial for obtaining reliable and reproducible models.⁶¹ This includes standardising pre-analytical and analytical steps, digital conversion, processing of final digital image files and their use in AI models. Ensuring the quality of WSIs – through precise dimension, orientation, colour and magnification – is paramount.⁶² A key step in AI development is the annotation of WSIs, which is time-consuming, subjective and requires expert involvement but remains crucial for AI training. The recently proposed active learning-based framework has shown promise in standardising this process, easing the burden of WSI labelling and reducing biases.⁶³

Another critical aspect is the transparent reporting of performance metrics during AI development, which helps assess models’ practical applicability and reproducibility. There is a gap in comparing AI performance with human evaluation, especially among non-experts, which would validate its usefulness in real-world pathology.⁴⁶ Additionally, randomised controlled trials (RCTs) are essential to establish the efficacy and safety of AI in assessing disease activity and predicting outcomes.⁶⁴ Using consistent and diverse datasets and ensuring external validation, particularly with publicly available datasets, is vital to ensure reproducibility and mitigate overfitting.

User-friendly AI interfaces are necessary to integrate AI into clinical practice and training programmes to familiarise clinicians and patients with AI. Creating models with trustworthy and easily understandable decision processes and outputs is crucial. Explainable AI is promising, providing interpretable results that make decision processes transparent and clearly show how specific histological features contribute to AI decisions.⁶⁵

Ethical concerns regarding data privacy, transparency and fairness also need to be addressed.^66,67 Robust cybersecurity measures are essential to safeguard personal data. It is vital to ensure clarity in model development and implement rigorous validation processes in collaboration with regulatory authorities to overcome ethical concerns related to transparency, accountability and the potential for discriminatory bias in decision-making.⁶⁸ Regular audits should be conducted to ensure fairness and prevent biases. Five principles, that is, beneficence, non-maleficence, autonomy, justice and accountability, should guide the ethical use of AI.⁶⁷

The cost-effectiveness of AI in IBD pathology is a critical factor for its integration into clinical practice.^69,70 Although the initial implementation costs are substantial, covering software, digital infrastructure, data storage and clinician training, AI offers promising long-term savings. By automating routine tasks and standardising diagnostic procedures, AI can reduce the workload on pathologists and increase throughput without additional staff. Furthermore, AI has the potential to decrease diagnostic errors, facilitate personalised medicine and enable early lesion detection, ultimately lowering long-term healthcare costs. However, the ongoing cost of maintaining and updating AI systems and integrating them into existing workflows requires careful consideration. Demonstrating the economic value of AI through RCTs is essential to provide concrete evidence of its benefits. Additionally, government and insurance funding support and collaborations between healthcare providers and AI developers are crucial to sharing and mitigating the financial burden of AI implementation. Particular attention should be given to the application of AI in resource-constrained settings, where high initial costs can be prohibitive and pose significant barriers, adding challenges to already overburdened healthcare systems.⁷¹ Validation of cost-effectiveness models is crucial, and regulatory authorities should consider provisions for reimbursement before integration into clinical practice. Additionally, leveraging open-access platforms, pre-trained models, fostering collaborative networks with well-resourced institutions and adopting an incremental implementation before full-scale rollout can make AI more accessible and cost-effective even in resource-constrained settings.^72,73

Future directions

While significant strides have been made in research on AI for histological analysis in IBD, a strategic focus on future directions and actionable steps is essential for maximising advancements in this field and facilitating its clinical implementation.

Prioritising funding initiatives to develop and validate AI models that consistently perform across diverse patient populations and healthcare settings is crucial for realising the full potential of this technology. Collaborative efforts among researchers, healthcare providers and industry experts to create large, high-quality and inclusive datasets are critical to overcoming current limitations in algorithm performance. A crucial aspect of this progress involves the expansion of data-sharing initiatives. Additionally, integrating histological AI analysis with multimodal data – such as endoscopic imaging, genetic profiles and clinical biomarkers – can provide a more comprehensive understanding of disease activity and progression, enabling personalised care. Real-time diagnostics represent another frontier for AI in histology. Advances in AI-powered tools that provide real-time pathology assessment during advanced endoscopic procedures could significantly reduce delays in care by enabling immediate decision-making and timely interventions.

The roadmap for integrating AI-aided digital pathology into clinical practice (Figure 2) involves key steps such as data standardisation, model transparency, external validation, explainability and ensuring privacy and fairness. The results of RCTs are eagerly anticipated, and collaboration between AI working groups and regulatory authorities is crucial to ensure the safe, reliable and equitable implementation of AI in IBD pathology.

Figure 2.

Roadmap to clinical implementation of AI-aided histology in IBD. This figure outlines the essential steps required to integrate AI-aided histology into clinical practice for IBD management. Each milestone on the roadmap corresponds to a crucial step, including data standardisation, model training and validation, transparent reporting, user-friendly integration, ethical and regulatory compliance and cost-effectiveness. This roadmap provides a structured approach for translating AI-powered histology into actionable, practical tools in IBD management, aiming to improve accuracy, efficiency and personalised patient care.

Conclusion

AI-aided digital pathology holds transformative potential in IBD, offering precise, consistent assessments of histologic features for disease activity, outcome prediction and therapeutic response in UC and CD. Its application improves diagnostic accuracy, standardises assessments and supports early lesion detection, advancing traditional methods in both clinical practice and trials. AI facilitates real-time analysis and adaptive designs in trials, aiding in dynamically informed treatment strategies. By predicting patient outcomes and treatment responses, AI supports a more personalised approach to care, and the integration of clinical, molecular and histological data expands its role in precision medicine.

Future research is crucial to fully harness AI’s potential in IBD. Developing robust, multimodal AI models requires large, diverse datasets and refining algorithms to integrate data sources like OMICs and endoscopic and histological findings. RCTs are needed to validate AI’s clinical efficacy, safety and cost-effectiveness alongside ethical considerations like data privacy and bias. With these advancements, AI could become integral to IBD management, promoting more effective, individualised care.

Footnotes

Appendix

Acknowledgements

None.

Declarations

ORCID iDs

Giovanni Santacroce

Olga Maria Nardone

Ivan Capobianco

Subrata Ghosh

Marietta Iacucci

References

GBD 2017 Inflammatory Bowel Disease Collaborators. The global, regional, and national burden of inflammatory bowel disease in 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet Gastroenterol Hepatol 2020; 5: 17–30.

Turner

Ricciuto

Lewis

, et al. STRIDE-II: an update on the selecting therapeutic targets in inflammatory bowel disease (STRIDE) initiative of the international organization for the study of IBD (IOIBD): determining therapeutic goals for treat-to-target strategies in IBD. Gastroenterology 2021; 160: 1570–1583.

Neurath

Vieth

. Different levels of healing in inflammatory bowel diseases: mucosal, histological, transmural, barrier and complete healing. Gut 2023; 72: 2164–2183.

Park

Abdi

Gentry

, et al. Histological disease activity as a predictor of clinical relapse among patients with ulcerative colitis: systematic review and meta-analysis. Am J Gastroenterol 2016; 111: 1692–1701.

Park

Kang

Yoon

, et al. Histologic evaluation using the Robarts histopathology index in patients with ulcerative colitis in deep remission and the association of histologic remission with risk of relapse. Inflamm Bowel Dis 2022; 28: 1709–1716.

Magro

Sabino

Rosini

, et al. ECCO position on harmonisation of Crohn’s disease mucosal histopathology. J Crohns Colitis 2022; 16: 876–883.

Magro

Doherty

Peyrin-Biroulet

, et al. ECCO position paper: harmonization of the approach to ulcerative colitis histopathology. J Crohns Colitis 2020; 14: 1503–1511.

Sands

Sandborn

Panaccione

, et al. Ustekinumab as induction and maintenance therapy for ulcerative colitis. N Engl J Med 2019; 381: 1201–1214.

D’Haens

Dubinsky

Kobayashi

, et al. Mirikizumab as induction and maintenance therapy for ulcerative colitis. N Engl J Med 2023; 388: 2444–2455.

10.

Peyrin-Biroulet

Bressenot

Kampman

. Histologic remission: the ultimate therapeutic goal in ulcerative colitis? Clin Gastroenterol Hepatol 2014; 12: 929–934.e2.

11.

Mosli

Feagan

Zou

, et al. Development and validation of a histological index for UC. Gut 2017; 66: 50–58.

12.

Marchal-Bressenot

Salleron

Boulagnon-Rombi

, et al. Development and validation of the Nancy histological index for UC. Gut 2017; 66: 43–49.

13.

Gui

Bazarova

Del Amor

, et al. PICaSSO Histologic Remission Index (PHRI) in ulcerative colitis: development of a novel simplified histological score for monitoring mucosal healing and predicting clinical outcomes and its applicability in an artificial intelligence system. Gut 2022; 71: 889–898.

14.

Geboes

Riddell

Ost

, et al. A reproducible grading scale for histological assessment of inflammation in ulcerative colitis. Gut 2000; 47: 404–409.

15.

Turner

Sinkre

Neumann

, et al. Primary colonic eosinophilia and eosinophilic colitis in adults. Am J Surg Pathol 2017; 41: 225–233.

16.

Morgenstern

Brook

Rinawi

, et al. Tissue and peripheral eosinophilia as predictors for disease outcome in children with ulcerative colitis. Dig Liver Dis 2017; 49: 170–174.

17.

Sedano

Almradi

, et al. An international consensus to standardize integration of histopathology in ulcerative colitis clinical trials. Gastroenterology 2021; 160: 2291–2302.

18.

Nardone

Iacucci

Villanacci

, et al. Real-world use of endoscopic and histological indices in ulcerative colitis: results of a global survey. United European Gastroenterol J 2023; 11: 514–519.

19.

Römkens

TEH

Kranenburg

Tilburg

A van

, et al. Assessment of histological remission in ulcerative colitis: discrepancies between daily practice and expert opinion. J Crohns Colitis 2018; 12: 425–431.

20.

Iacucci

Santacroce

Zammarchi

, et al. Artificial intelligence and endo-histo-omics: new dimensions of precision endoscopy and histology in inflammatory bowel disease. Lancet Gastroenterol Hepatol 2024; 9: 758–772.

21.

Matalka

Al-Omari

Salama

, et al. A novel approach for quantitative assessment of mucosal damage in inflammatory bowel disease. Diagn Pathol 2013; 8: 156.

22.

Vande Casteele

Leighton

Pasha

, et al. Utilizing deep learning to analyze whole slide images of colonic biopsies for associations between eosinophil density and clinicopathologic features in active ulcerative colitis. Inflamm Bowel Dis 2022; 28: 539–546.

23.

Kiyokawa

Abe

Matsui

, et al. Deep learning analysis of histologic images from intestinal specimen reveals adipocyte shrinkage and mast cell infiltration to predict postoperative Crohn disease. Am J Pathol 2022; 192: 904–916.

24.

del Amor

Meseguer

Parigi

, et al. Constrained multiple instance learning for ulcerative colitis prediction using histological images. Comput Methods Programs Biomed 2022; 224: 107012.

25.

Ohara

Nemoto

Maeda

, et al. Deep learning-based automated quantification of goblet cell mucus using histological images as a predictor of clinical relapse of ulcerative colitis with endoscopic remission. J Gastroenterol 2022; 57: 962–970.

26.

Najdawi

Sucipto

Mistry

, et al. Artificial intelligence enables quantitative assessment of ulcerative colitis histology. Mod Pathol 2023; 36: 100124.

27.

Rymarczyk

Schultz

Borowa

, et al. Deep learning models capture histological disease activity in Crohn’s disease and ulcerative colitis with high fidelity. J Crohns Colitis 2024; 18(4): 604–614.

28.

Iacucci

Parigi

Del Amor

, et al. Artificial intelligence enabled histological prediction of remission or activity and clinical outcomes in ulcerative colitis. Gastroenterology 2023; 164: 1180–1188.e2.

29.

Liu

Prasath

Siddiqui

, et al. Machine learning-based prediction of pediatric ulcerative colitis treatment response using diagnostic histopathology. Gastroenterology 2024; 166: 921–924.e4.

30.

Wang

Yao

, et al. Histological image-based ensemble model to identify myenteric plexitis and predict endoscopic postoperative recurrence in Crohn’s disease: a multicentre, retrospective study. J Crohns Colitis 2024; 18: 727–737.

31.

Peyrin-Biroulet

Adsul

Stancati

, et al. An artificial intelligence-driven scoring system to measure histological disease activity in ulcerative colitis. United European Gastroenterol J 2024; 12: 1028–1033.

32.

Furlanello

Bussola

Merzi

, et al. The development of artificial intelligence in the histological diagnosis of inflammatory bowel disease (IBD-AI). Dig Liver Dis 2025; 57(1): 184–189.

33.

Ohara

Maeda

Ogata

, et al. Automated neutrophil quantification and histological score estimation in ulcerative colitis. Clin Gastroenterol Hepatol. Epub ahead of print July 2024. DOI: 10.1016/j.cgh.2024.06.040.

34.

Reigle

Lopez-Nunez

Drysdale

, et al. Using deep learning to automate eosinophil counting in pediatric ulcerative colitis histopathological images. medRxiv. Epub ahead of print April 2024. DOI: 10.1101/2024.04.03.24305251.

35.

Rubin

Kubassova

Weber

, et al. Deployment of an artificial intelligence histology tool to aid qualitative assessment of histopathology using the Nancy histopathology index in ulcerative colitis. Inflamm Bowel Dis. Epub ahead of print September 2024. DOI: 10.1093/ibd/izae204.

36.

Puga-Tejada

Majumder

Maeda

, et al. Artificial intelligence-enabled histology exhibits comparable accuracy to pathologists in assessing histological remission in ulcerative colitis: a systematic review, meta-analysis and meta-regression. J Crohns Colitis 2025; 19(1): jjae198.

37.

Magro

Pai

Kobayashi

, et al. Resolving histological inflammation in ulcerative colitis with mirikizumab in the LUCENT induction and maintenance trial programmes. J Crohns Colitis 2023; 17: 1457–1470.

38.

Pai

D’Haens

Kobayashi

, et al. Histologic assessments in ulcerative colitis: the evidence behind a new endpoint in clinical trials. Expert Rev Gastroenterol Hepatol 2024; 18: 73–87.

39.

Gottlieb

Daperno

Usiskin

, et al. Endoscopy and central reading in inflammatory bowel disease clinical trials: achievements, challenges and future developments. Gut 2021; 70(2): 418–426.

40.

Balzer

Cai

Godoy Garraza

, et al. Adaptive selection of the optimal strategy to improve precision and power in randomized trials. Biometrics 2024; 80(1): ujad034.

41.

Iacucci

Santacroce

Meseguer

, et al. MP383 novel AI-enabled assessment of histologic remission and early response to therapy in a phase 2 clinical trial UC cohort. United European Gastroenterol J 2024; 12: 439.

42.

Harun

Kassir

, et al. Machine Learning-based quantification of patient factors impacting remission in patients with ulcerative colitis: insights from etrolizumab phase III clinical trials. Clin Pharmacol Ther 2024; 115: 815–824.

43.

Takishima

Maeda

Ogata

, et al. Beyond complete endoscopic healing: Goblet appearance using an endocytoscope to predict future sustained clinical remission in ulcerative colitis. Dig Endosc 2022; 34: 1030–1039.

44.

Yoon

Jangi

Dulai

, et al. Incremental benefit of achieving endoscopic and histologic remission in patients with ulcerative colitis: a systematic review and meta-analysis. Gastroenterology 2020; 159: 1262–1275.e7.

45.

Dekker

Nass

Iacucci

, et al. Performance measures for colonoscopy in inflammatory bowel disease patients: European Society of Gastrointestinal Endoscopy (ESGE) Quality Improvement Initiative. Endoscopy 2022; 54: 904–915.

46.

Maeda

Kudo

Santacroce

, et al. Artificial intelligence-assisted colonoscopy to identify histologic remission and predict the outcomes of patients with ulcerative colitis: a systematic review. Dig Liver Dis 2024; 56: 1119–1125.

47.

Ahmad

East

Panaccione

, et al. Artificial intelligence in inflammatory bowel disease: implications for clinical practice and future directions. Intest Res 2023; 21: 283–294.

48.

Iacucci

Jeffery

Acharjee

, et al. Computer-aided imaging analysis of probe-based confocal laser endomicroscopy with molecular labeling and gene expression identifies markers of response to biological therapy in IBD patients: the Endo-Omics Study. Inflamm Bowel Dis 2023; 29: 1409–1420.

49.

Fei

Gao

, et al. Towards artificial general intelligence via a multimodal foundation model. Nat Commun 2022; 13: 3094.

50.

Moor

Banerjee

Abad

ZSH

, et al. Foundation models for generalist medical artificial intelligence. Nature 2023; 616: 259–265.

51.

Howell

Corrado

DeSalvo

. Three epochs of artificial intelligence in health care. JAMA 2024; 331: 242.

52.

Rath

Atreya

Bodenschatz

, et al. Intestinal barrier healing is superior to endoscopic and histologic remission for predicting major adverse outcomes in inflammatory bowel disease: the Prospective ERIca Trial. Gastroenterology 2023; 164: 241–255.

53.

Zammarchi

Santacroce

Puga-Tejada

, et al. Epithelial neutrophil localization and tight junction Claudin-2 expression are innovative outcome predictors in inflammatory bowel disease. United European Gastroenterol J 2024; 12(9): 1155–1166.

54.

Iacucci

Santacroce

Majumder

, et al. Opening the doors of precision medicine: novel tools to assess intestinal barrier in inflammatory bowel disease and colitis-associated neoplasia. Gut 2024; 73: 1749–1762.

55.

Majumder

Santacroce

Maeda

, et al. Endocytoscopy with automated multispectral intestinal barrier pathology imaging for assessment of deep healing to predict outcomes in ulcerative colitis. Gut 2024; 73: 1603–1606.

56.

Sahoo

Swanson

Sayed

, et al. Artificial intelligence guided discovery of a barrier-protective therapy in inflammatory bowel disease. Nat Commun 2021; 12: 4246.

57.

Abdelrahim

Siggens

Iwadate

, et al. New AI model for neoplasia detection and characterisation in inflammatory bowel disease. Gut 2024; 73: 725–728.

58.

Bera

Schalper

Rimm

, et al. Artificial intelligence in digital pathology – new tools for diagnosis and precision oncology. Nat Rev Clin Oncol 2019; 16: 703–715.

59.

Kobayashi

Tomita

Shimizu

, et al. P53 Expression as a diagnostic biomarker in ulcerative colitis-associated cancer. Int J Mol Sci 2017; 18: 1284.

60.

Noguchi

Ando

Emoto

, et al. Artificial intelligence program to predict p53 mutations in ulcerative colitis-associated cancer or dysplasia. Inflamm Bowel Dis 2022; 28: 1072–1080.

61.

Niazi

MKK

Parwani

Gurcan

. Digital pathology and artificial intelligence. Lancet Oncol 2019; 20: e253–e261.

62.

Basak

Ozyoruk

Demir

. Whole slide images in artificial intelligence applications in digital pathology: challenges and pitfalls. Turk Patoloji Derg 2023; 39: 101–108.

63.

Santacroce

Meseguer

Zammarchi

, et al. P406 a novel active learning-based digital pathology protocol annotation for histologic assessment in ulcerative colitis using PICaSSO histologic remission index (PHRI). J Crohns Colitis 2024; 18: i843–i844.

64.

Han

Acosta

Shakeri

, et al. Randomised controlled trials evaluating artificial intelligence in clinical practice: a scoping review. Lancet Digit Health 2024; 6: e367–e373.

65.

van der Velden

BHM

Kuijf

Gilhuijs

KGA

, et al. Explainable artificial intelligence (XAI) in deep learning-based medical image analysis. Med Image Anal 2022; 79: 102470.

66.

Heidt

. Intellectual property and data privacy: the hidden risks of AI. Nature. Epub ahead of print September 2024. DOI: 10.1038/d41586-024-02838-z.

67.

Floridi

Cowls

. A unified framework of five principles for AI in society. Harv Data Sci Rev. Epub ahead of print June 2019. DOI: 10.1162/99608f92.8cd550d1.

68.

Ibrahim

Liu

Denniston

. Reporting guidelines for artificial intelligence in healthcare research. Clin Exp Ophthalmol 2021; 49: 470–476.

69.

Areia

Mori

Correale

, et al. Cost-effectiveness of artificial intelligence for screening colonoscopy: a modelling study. Lancet Digit Health 2022; 4: e436–e444.

70.

Youssef

Nichol

Martinez-Martin

, et al. Ethical considerations in the design and conduct of clinical trials of artificial intelligence. JAMA Netw Open 2024; 7: e2432482.

71.

Anirvan

Meher

Singh

. Artificial intelligence in gastrointestinal endoscopy in a resource-constrained setting: a reality check. Euroasian J Hepatogastroenterol 2020; 10: 92–97.

72.

Zehra

Parwani

Abdul-Ghafar

, et al. A suggested way forward for adoption of AI-enabled digital pathology in low resource organizations in the developing world. Diagn Pathol 2023; 18: 68.

73.

Zehra

Marino

Wang

, et al. Rethinking histology slide digitization workflows for low-resource settings. Med Image Comput Comput Assist Interv 2024; 15004: 427–436.