AI tested to automate biopsy analysis for coeliac diagnosis in Cambridge, UK

There's a quiet revolution stirring in Cambridge, and it could make life a whole lot easier for people living with coeliac disease. A homegrown AI tool, built by Lyzeum Ltd — a University of Cambridge spinout — is being trained to fast-track biopsy analysis. Backed by funding from the National Institute for Health and Care Research, this project is all about slashing diagnosis times and giving patients quicker answers.

As Professor Liz Soilleux, who leads the project, puts it: the algorithm's already hitting the mark, nailing correct diagnoses "in 97 out of 100 cases." Not too shabby for a machine.

How does it work?

The AI's been fed a rich diet of data — 4,000 biopsy images from five NHS hospitals — to spot the subtle differences between healthy tissue and signs of coeliac disease.

Then came the real test: another 650 biopsy samples. The AI flexed its digital muscles, pulling off a 97% accuracy rate.

What's clever here is how the tool learns to read these images like a seasoned pathologist, picking out patterns that aren't obvious to the human eye. But there's a hitch: patients are wary about how AI reaches its decisions. CUH researchers stress that explainability is a must — people want to know why a diagnosis is made, not just what it is.

As CUH puts it: "Ensuring diagnoses are explainable is key for the researchers and is likely to be a critical step in the AI being approved, and trusted, for use across the NHS."

Florian Jaeckle, the study's first author, says they're now ramping up tests with a much bigger batch of samples. "That'll put us in a position to share this device with the regulator," he explains, "bringing us nearer to this tool being used in the NHS."

Why does it matter?

Speed matters. For anyone waiting on a coeliac diagnosis, delays can feel endless — we're talking months, sometimes longer. Faster results could mean faster treatment, and in turn, a much better quality of life.

• Coeliac disease affects roughly 1 in 100 people in the UK.
• Many remain undiagnosed, suffering silently while awaiting confirmation.

AI has the potential to change that. It's not about replacing doctors but giving them sharper tools for the job. Plus, it could ease the strain on overstretched NHS pathology services.

And there's a bigger picture too. As part of a broader NHS shift towards digital diagnostics, this tech signals a future where AI lends a hand across the board — from spotting cancer early to decoding dizziness.

The context

The UK government's already splashing serious cash into AI-driven healthcare, pledging £82.6 million for research across three big projects. Two of these are laser-focused on using AI to tackle cancer head-on.

Take PharosAI, for example:

• It's bringing together NHS and Biobank data on one powerful, secure platform.
• The goal? Build AI models to crack tough medical riddles, like cancer diagnosis and treatment breakthroughs.

There's also plenty of other AI work buzzing around the NHS:

• University Hospitals Dorset and Bournemouth University are hunting for volatile organic compounds (VOCs) linked to skin cancer.
• Over in Norfolk and Norwich, researchers are trialling CAVA, a device that tracks eye and head movements to figure out causes of dizziness. Twenty hospitals are already on board.

All these efforts point in one direction: a healthcare system that's smarter, faster, and more precise. Cambridge's coeliac project is just one bright thread in this fast-weaving tapestry.