Getting diagnosed with coeliac disease can be a long and frustrating process, but a new artificial intelligence (AI) tool developed by scientists at the University of Cambridge might soon change that.
In a recent study, this AI model was able to correctly identify whether a person had coeliac disease in 97 out of 100 cases by analyzing images from their gut biopsies.
Coeliac disease is an autoimmune condition that is triggered when people eat gluten, a protein found in wheat, barley, and rye. When someone with this disease eats gluten, their immune system mistakenly attacks the lining of the small intestine.
This can lead to a wide range of symptoms, including stomach pain, diarrhea, tiredness, weight loss, skin rashes, and low iron levels. Because symptoms can vary so much between individuals, doctors often struggle to make a clear diagnosis.
Currently, the best way to diagnose coeliac disease is to take a biopsy from the small intestine, specifically the duodenum. A specialist then examines the sample under a microscope to look for damage to tiny finger-like structures called villi.
These villi are essential for absorbing nutrients. If they are damaged or flattened, it’s a strong sign of coeliac disease. However, judging this damage can be tricky.
Different specialists may see the same sample differently, especially when the changes are small. To make it more consistent, doctors use a scale called the Marsh-Oberhuber classification, which ranks the damage from zero (no damage) to four (severe damage).
The team at Cambridge wanted to see if AI could help make this process easier and more reliable. They trained their algorithm using more than 4,000 biopsy images collected from five hospitals. These images were taken using different equipment and methods to make sure the AI could handle a variety of real-world conditions.
The results were impressive. When tested on a new set of 650 biopsy images, the AI correctly diagnosed coeliac disease with over 95% sensitivity and nearly 98% specificity. In simple terms, it was very good at telling who did and didn’t have the disease.
This performance is as good as or even better than experienced doctors. In fact, when four pathologists reviewed 30 slides, they agreed with the AI just as often as they agreed with each other.
The research team believes that this kind of AI tool could be very helpful in hospitals, especially where doctors are overworked or where there are not enough specialists—such as in developing countries. Faster and more reliable diagnoses could help patients start treatment sooner and ease the burden on healthcare systems.
The scientists are now working on testing the algorithm with even larger groups of patients. Their goal is to eventually get approval for use in the UK’s National Health Service (NHS). They’ve also started a company called Lyzeum Ltd to help bring this technology to market.
The study was published in the journal NEJM AI and funded by Coeliac UK, Innovate UK, and other health research organizations.
Coeliac UK’s research officer, Keira Shepherd, highlighted the importance of quick diagnosis, as patients must continue eating gluten during the testing period, which can cause painful symptoms. This research could shorten that difficult waiting time.
One challenge the team is thinking about is helping people understand how the AI works. Patients and doctors want to know not just the diagnosis, but how the machine made its decision. Building public trust in AI tools will require clear explanations and transparency.
In conclusion, this study shows that AI has the power to transform how we diagnose coeliac disease. It could mean faster results, fewer errors, and better care for patients around the world.
If you care about gut health, please read studies about a surprising link between gut health and eye disease risk and Gut troubles could signal future Parkinson’s disease.
For more information about gut health, please read studies that this stuff in coffee and chocolate may impact gut health and Gut health is linked to anxiety levels.
The study is published in NEJM AI.
Copyright © 2025 Knowridge Science Report. All rights reserved.
