An artificial intelligence tool developed by researchers at the University of Cambridge, University College London, and Queen Mary University of London has demonstrated improved accuracy in identifying abnormalities in blood cells compared to human experts. The system, called CytoDiffusion, uses generative AI technology similar to that behind image generators such as DALL-E to analyze the shape and structure of blood cells.
The study was published in the journal Nature Machine Intelligence. According to Simon Deltadahl from Cambridge’s Department of Applied Mathematics and Theoretical Physics and first author of the study, “We’ve all got many different types of blood cells that have different properties and different roles within our body. White blood cells specialise in fighting infection, for example. Knowing what an unusual or diseased blood cell looks like under a microscope is an important part of diagnosing many diseases.”
A typical blood smear contains thousands of cells, making it difficult for humans to analyze every cell. Deltadahl noted, “Humans can’t look at all the cells in a smear – it’s just not possible. Our model can automate that process, triage the routine cases, and highlight anything unusual for human review.”
Dr Suthesh Sivapalaratnam from Queen Mary University of London described his experience as a junior haematology doctor: “The clinical challenge I faced as a junior haematology doctor was that after a day of work, I would have a lot of blood films to analyse. As I was analysing them in the late hours, I became convinced AI would do a better job than me.”
CytoDiffusion was trained on over half a million images collected at Addenbrooke’s Hospital in Cambridge. This dataset included both common and rare cell types as well as elements that often confuse automated systems. By modeling the full range of cell appearances rather than just separating categories, CytoDiffusion became more robust to variations between hospitals, microscopes, and staining methods.
Testing showed that CytoDiffusion detected abnormal cells linked to leukaemia with higher sensitivity than existing systems. It also performed as well or better than current models even with fewer training examples and could quantify its own uncertainty. Deltadahl said, “When we tested its accuracy, the system was slightly better than humans. But where it really stood out was in knowing when it was uncertain. Our model would never say it was certain and then be wrong, but that is something that humans sometimes do.”
Professor Michael Roberts from Cambridge added: “We evaluated our method against many of the challenges seen in real-world AI, such as never-before-seen images, images captured by different machines and the degree of uncertainty in the labels. This framework gives a multi-faceted view of model performance, which we believe will be beneficial to researchers.”
The team also found that CytoDiffusion could generate synthetic images indistinguishable from real ones; ten experienced haematologists were unable to reliably tell real from AI-generated images during testing. “That really surprised me,” said Deltadahl. “These are people who stare at blood cells all day, and even they couldn’t tell.”
Researchers are releasing what they describe as the world’s largest publicly available dataset of peripheral blood smear images—over half a million—to support further development worldwide.
“By making this resource open, we hope to empower researchers worldwide to build and test new AI models, democratise access to high-quality medical data, and ultimately contribute to better patient care,” said Deltadahl.
Despite these advances, researchers emphasize that CytoDiffusion is not intended as a replacement for clinicians but rather as support for rapidly flagging abnormal cases while automating routine analysis.
Professor Parashkev Nachev from UCL stated: “The true value of healthcare AI lies not in approximating human expertise at lower cost, but in enabling greater diagnostic, prognostic, and prescriptive power than either experts or simple statistical models can achieve... This ‘metacognitive’ awareness – knowing what one does not know – is critical to clinical decision-making, and here we show machines may be better at it than we are.”
Further work is planned to improve speed and test CytoDiffusion across diverse populations.
Funding for this research came from several organizations including Trinity Challenge; Wellcome; British Heart Foundation; Cambridge University Hospitals NHS Foundation Trust; Barts Health NHS Trust; NIHR Cambridge Biomedical Research Centre; NIHR UCLH Biomedical Research Centre; NHS Blood and Transplant; with research conducted by the Imaging working group of BloodCounts! consortium.
Simon Deltadahl is affiliated with Lucy Cavendish College at Cambridge.