AI Aids in Assessment of Ulcerative Colitis Activity, Remission

Not only are artificial intelligence (AI) systems potentially highly accurate for assessment of disease activity and remission of ulcerative colitis (UC), but they can mitigate some limits of human assessment, according to presentations at the 17th congress of the European Crohn's and Colitis Organisation.

Importantly, AI systems have the potential to supplement the services of expert histopathologists and endoscopists rather than replace them, several experts asserted at the meeting.

"We will always need pathologists," reassured inflammatory bowel disease (IBD) specialist Laurent Peyrin-Biroulet, MD, PhD, of Nancy (France) University Hospital, who presented about the use of an AI-driven scoring system to measure histological disease activity in UC.

Peyrin-Biroulet, who is the president of ECCO and acts as the scientific secretary of the International Organization for the Study of IBD, added that the use of AI systems could mean that pathologists have more time to do other tasks. Not only that, but it's also not always possible to have IBD pathologist in every center, everywhere in the world.

"If we can get something that will automatically evaluate the disease activity, I think it will be something fantastic," Peyrin-Biroulet said, "and it's the reason why we were thinking that there is a need for an automated method to measure histological activity in UC."

Old Concept Enhancing Current Practice

The idea of using AI systems to aid diagnostics is not new but now makes even more sense in the post–COVID-19 era, suggested Aaron F. Pollett, MD, MSc, FRCPC, codirector of the division of diagnostic medical genetics at Mount Sinai Hospital in Toronto and a pathologist with a specialty interest in gastrointestinal pathology.

"When we talk about artificial intelligence and histology, there's actually a very long history, it goes back over 30 years," Pollett said, from assessing cervical samples to its use in breast screening.

What seems to be sudden flurry of activity in the world of AI and pathology in recent years comes down to having a higher capacity for looking at large images, having access to large data sets, and having a high amount of computing power, Pollet inferred. Moreover, "the capacity and the need for whole slide imaging has really grown especially in the last few years as the pandemic has forced centers to adopt." The need to work remotely and flexibly across centers and the number of available pathologists have also played a role.

AI systems that use image-based retrieval systems are making good headway in IBD, particularly in the diagnosis of UC where "some of the initial research is showing it can be quite good," said Pollett. The "patchiness that Crohn's can have in comparison to UC" means that it's still an emerging area, but can perhaps be useful for more questionable cases in which "having that degree of certainty can certainly help because there is a discrepancy between specialist and nonspecialist pathologists in the likelihood that what they predict on the biopsy will be the underlying disease."

AI Systems in IBD – Do They Work?

Histopathology is becoming increasingly integrated into IBD clinical trial design at the behest of the Food and Drug Administration and European associations such as ECCO. This can be a tedious procedure that can be prone to error and disagreement between scorers.

The AI-driven scoring system that Peyrin-Biroulet and associates have been working on aims to fix all that by using machine learning and image processing to set up a reproducible system. Their system, which is based on the Nancy histological index for UC, shows high correlation (87%) with histopathologists' assessment and was 100% accurate in identifying images with high (grade 4) or no (grade 0) inflammatory activity. The accuracy decreased, however, when trying to distinguish between more moderate activity, with a 75% accuracy for identifying grade 3 and 82% accuracy for grades 1 or 2.

"I'm actually very fascinated to see how we can be supported by the AI work in our practice," observed Francesca Rosini, a histopathologist working at S. Orsola–Malpighi University Hospital in Bologna, Italy.

Rosini, who chaired the digital oral presentation session in which Peyrin-Biroulet had presented also noted that "obviously for us as well [as AI systems] no activity or severe activity is the easiest part but when it's in between that's where the problems come."

Simplifying Histological Scoring

Simplifying scoring for use in AI systems could be the key to their future success, as Tommaso Lorenzo Parigi, MD, from Humanitas University in Milan, and a research fellow at the University of Birmingham (England), suggested.

"Histology is particularly important to distinguish between mild activity and remission," Parigi said. "More than 30 histological scores that have been proposed, but their adoption in clinical practice remains limited."

Parigi has been part of an international team that has developed a simplified histological score based on "the presence of absence of neutrophils, regardless of their number," since these are "key determinants of disease activity".

The score, known as the Paddington International Virtual Chromoendoscopy Scre (PICaSSO) Histologic Remission Index (PHRI), has been shown to correlate well with endoscopic outcomes and thus a good measure to include in AI systems. The results of this work were published online in Gut to coincide with the ECCO congress.

"We are getting close to a world where we could screen biopsies with this kind of systems and consider skipping the pathologists result if AI detected activity," Parigi provocatively suggested. "Of course, we need to increase and improve our sensitivity, and we are currently working on that to reduce false negatives, as well as training our model to use and apply other histological scores."

Assessing the Gut in Real Time

Perhaps one of the most exciting developments it to be able to use these AI technologies to examine the gut in real time.

"Virtual chromoendoscopy will give you the opportunity to distinguish very carefully all the details of mucosal vascular pattern," said Marietta Iacucci MD, PhD, FASGE, AGAF, an associate professor and gastroenterology consultant at the Birmingham (England) University Hospitals.

"So AI can give you, in real time, the score but at the same time it can help to target, to do biopsies for healing," Iacucci added when reporting the results of a study evaluating the performance of the first virtual chromoendoscopy AI system to detect endoscopic and histologic remission in UC.

The system was proven to predict endoscopic remission very accurately (94% using PICaSSO and 87% using the UC endoscopic index of severity) when compared with a human endoscopist. Rates of predicting histological remission were also high, at around 83%-85%, depending on the score used.

"For the future, this AI tool can expediate, support, and standardize the endoscopic evaluation of UC mucosal healing in clinical practice and in clinical trials," Iacucci said.

The next steps are to combine virtual chromoendoscopy with the PHRI and to validate the tool in a multicenter, international PICaSSO-AI study.

The AI-driven scoring system presented by Peyrin-Biroulet was supported by Takeda. Peryin-Biroulet acknowledged the receipt of personal fees and grants from Takeda along with multiple other Pharma companies and owning stock options from CTMA. Iacucci has received research grants from Pentax, AbbVie, Olympus, and Fujifilm and personal fees from Pentax, AbbVie and Janssen. Pollett, Rosini, and Parigi had no financial conflicts of interest to disclose.

This article originally appeared on MDedge.com, part of the Medscape Professional Network.

Comments

Commenting is limited to medical professionals. To comment please Log-in.

Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.