In June 2023, SciCom selected a joint proposal from Shaji Sebastian (then a ClinCom Member, now a Governing Board Member) and the H-ECCO Chair, Aart Mookhoek, addressing the present and future directions of artificial intelligence (AI) in IBD at the frontiers of imaging (endoscopy, radiology, histology), big data, predictive models, drug discovery and both ethical and regulatory considerations.
After a call for participants, a workshop brought together 23 emerging experts in the field of AI in IBD, including gastroenterologists, pathologists, surgeons and a geneticist, representing different committees within ECCO, including ClinCom, H-ECCO, Y-ECCO, EduCom and S-ECCO, to prepare the first ECCO papers on AI. The rapidly emerging field covers many facets and four papers will be published in JCC over the coming months, namely: AI and Imaging, Medical and Precision Medicine, Surgical Management, and Regulatory and Methodological Considerations.
These papers will provide valuable information, identifying areas where AI might deliver benefit now as well as hurdles that will need to be overcome in the future.
For example, in the field of radiology, AI can outperform radiologists for difficult tasks, such as IBD subtyping, detection of subtle inflammatory activity and prediction of fibrosis. There are efforts to combine radiomic features with other data, such as clinical and biochemical data, to provide more prognostic power than is currently available. Yet, while institutes are using different imaging protocols, it is difficult to use data from another institute to validate an algorithm.
For endoscopy, the majority of the literature in the field focuses on disease activity scoring. Use of an Al algorithm can reduce diagnostic variability among endoscopists. Ideally, we will be integrating endoscopy data with other data sources such as histology and further omics data. The product of such an endo-histo model could revolutionise personalised healthcare [1].
In histology, the AI algorithms match expert pathologists using common scoring systems such as the Nancy index and the Robarts Histopathology Index. These scores solve the problem of inter-observer heterogeneity. Moreover, there is a worldwide shortage of pathologists, so AI applications may reduce the workload for them.
We can also use algorithms to unravel IBD pathophysiology. A recent study described an AI model to predict postoperative outcome after surgery in patients with CD [2]. The algorithm was shown to have learned that there is an association between adipocyte morphology and likelihood of relapse [2].
Again, the ideal is to combine data sets from genomics, transcriptomics, the metabolome and the microbiome. For example, a random forest method may be used to select features from the metabolome and microbiome together or a deep neural network may be employed to assess the proteome and the transcriptome in patients with CD [3]. AI is very helpful in making sense of these large data sets, for example through the identification of differentially expressed proteins and genes, and the hope is that this will lead to the discovery of biomarkers predictive of therapeutic success. Nevertheless, such technologies are resource intensive and to date only limited sample sizes have been used for these studies, leading to a tendency towards overfitting and limited success in validating the biomarkers.
Advances in monitoring via the use of wearables for patients with IBD are impressive; for example, a new device [4] can predict flares up to 20 days in advance by measuring nocturnal awakenings and respiratory rate, and a sweat sensor [5] can be used to measure TNF-alpha levels accurately, a very high correlation having been found between sweat and serum TNF-alpha levels. Wearables are indeed “cool”, but enthusiasm quickly wanes and attrition rates are a problem. Perhaps there will be more incentive to use wearables to predict a flare once it can also be prevented.
For the next generation of surgeons, AI promises great benefits in reducing surgical time and costs. This is especially true in robotic surgery, where we can combine performance metrics with surgical phase recognition to provide AI feedback to surgeons in training [6, 7].
AI is replacing classic statistical techniques used in the field of surgical research, and large language models are now able to extract data from electronic medical records and patient-reported outcomes from online fora [8]. As regards to our IBD patient population, we could develop virtual cohorts [9] or automatically select patients eligible for inclusion in a clinical trial. One may, however, need to think about the ethical considerations of doing this, and it is important to understand the regulations and rules which come with AI.
In fact, within medicine all AI applications are considered a medical device and come with their own set of regulations and rules. The Hiroshima Process [10], signed by many countries, provides guiding principles and a code of conduct for the development of AI algorithms. They focus on patient outcomes and explainable AI (XAI). XAI [11] in practice may mean that an AI algorithm reports a score together with the level of certainty. This builds confidence for both patient and healthcare practitioner. Another example of XAI is when the algorithm highlights which part(s) of the image is used to make a certain decision.
The future of AI applications in IBD is bright, even when accounting for the resources required to run AI platforms, the proprietary platform software and the substantial training costs [12]. There is now a significant opportunity for ECCO to show leadership and support IBD specialists as they navigate between the hype and the practical knowledge needed to optimise the use of AI. ECCO could create a framework on how to design a study using AI, with guidelines and standardisation. There is certainly a need for more work to overcome the challenges. We strongly encourage you to read the four AI in IBD manuscripts in JCC for more inspiration.
On behalf of the SWS9 Steering Committee, many thanks to the authors of and contributors to this remarkable manuscript collaboration. If you would like to contribute to the next Scientific Workshop, the deadline for proposals in May 31, 2025.
*What is an ECCO Scientific Workshop? The ECCO Scientific Workshop is a biennial manuscript project, led by SciCom, that takes one under-researched topic and identifies research gaps and sets research questions for external investigators to address. Learn more here:
Call for Scientific Workshop 10 Topics – deadline for proposals May 31, 2025
