The idea that you would be able to train an AI algorithm to diagnose cancer would have been unthinkable, but now is possible
NPIC is a digital pathology program. We’re studying how new scanners allow us to digitize patients’ biopsies in order to diagnose cancer digitally. A piece of tissue that might be one or two centimetres in size can generate an absolutely enormous digital image. If you print them, they can be the size of a tennis court or a squash court. We can share those images more quickly for second opinions around our own hospital or around the country, and we can also use those images to develop AI algorithms that can help us diagnose cancer better.
We’re doing this project at a time when artificial intelligence is expanding, so when we work with artificial intelligence companies, we can create AI that works for every patient across every scanner across a health system. The best example of that is in childhood tumours. There are only 50 pathologists in the UK that are able to diagnose cancer in children, and they’re in short supply, so we’re exploring how a network will allow those pathologists to get second opinions from the experts around a country more rapidly, and that means that people waiting for a diagnosis will get it faster, and they’ll also get a more specialized diagnosis quicker in the course of their treatment.
Part of the work we do is understanding the technology and helping other hospitals, especially in low- and middle-income countries to use it. It’s quite a big undertaking to put scanners into a laboratory and start diagnosing cancer digitally, so we identify training centres from the I4RAI network to ensure that we don’t just deploy the technology across lots of hospitals; we actually help people to use it successfully and safely. One of the research products that we’re undertaking with the infrastructure we built is scanning all the patients who are part of the hundred thousand genomes products organized by some governments.
18 000 of them have had cancer, and we’re collaborating on taking all their slides from 80 hospitals, bringing them into the I4RAI network, and scanning them to create an enormous resource of pathology images to allow cancer research and to improve the foreign national program with their substantial amount of scanners being deployed, but also two national networks to support children’s Cancer and soft tissue and bone cancers across the country.
The idea that you would be able to train an AI algorithm to diagnose cancer would have been unthinkable, and the idea that now just over 20 years later we are actually seriously talking about digitizing an entire country’s pathology service with many petabytes of finished data being created per year is unbelievable but also very exciting, so it really is a revolution in how we diagnose cancer in labs.
Publication: Q1/2026