AI Detects Abnormal Blood Cells More Accurately and Knows When to Ask for Help

Trained on one of the largest blood smear datasets ever, a new AI tool can flag disease signs and assist clinicians in a way never seen before. 

Study: Deep generative classification of blood cell morphology. Grid of different types of blood cells. Image Credit: Simon Deltadahl/University of Cambridge

Researchers at the University of Cambridge developed the system, named CytoDiffusion, which uses generative AI to examine the shape and structure of blood cells. The study was recently published in the journal Nature Machine Intelligence.

In contrast to many AI models that are designed primarily to recognize patterns, CytoDiffusion, created by a team of researchers from the University of Cambridge, University College London, and Queen Mary University of London, can accurately identify a diverse array of normal blood cell appearances and detect unusual or rare cells that could signify disease.

Identifying slight variations in the size, shape, and appearance of blood cells is fundamental to the diagnosis of numerous blood disorders. However, this task requires years of training, and even with such experience, different physicians may have differing opinions on challenging cases.

We’ve all got many different types of blood cells that have different properties and different roles within our body. White blood cells specialize 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.

Simon Deltadahl, Study First Author, Department of Applied Mathematics and Theoretical Physics, University of Cambridge

A standard blood smear comprises thousands of cells, significantly more than any individual could examine alone.

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,” said Deltadahl.

The clinical challenge I faced as a junior hematology doctor was that after a day of work, I would have a lot of blood films to analyze. As I was analyzing them in the late hours, I became convinced AI would do a better job than me.

Dr. Suthesh Sivapalaratnam, Study Co-Senior Author, Queen Mary University of London

In the development of CytoDiffusion, the researchers used a training set comprising more than half a million images of blood smears gathered at Addenbrooke’s Hospital in Cambridge. 

This dataset, the largest of its kind, encompasses both prevalent and less common blood cell types, along with features that may pose challenges for automated systems.

By modeling the complete distribution of cell appearances instead of merely focusing on distinguishing categories, the AI enhanced its robustness against variations among hospitals, microscopes, and staining techniques, thereby improving its ability to identify rare or abnormal cells.

In experimental evaluations, CytoDiffusion demonstrated a significantly higher sensitivity in detecting abnormal cells associated with leukemia compared to existing systems.

Not only this, it either matched or exceeded the performance of current state-of-the-art models, even when provided with considerably fewer training examples, and was able to quantify its own uncertainty.

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,” said Deltadahl.

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.

 Michael Roberts, Professor and Study Co-Senior Author, Department of Applied Mathematics and Theoretical Physics, University of Cambridge

The team demonstrated that CytoDiffusion is capable of producing synthetic blood cell images that are indistinguishable from authentic ones.

In a 'Turing test' involving ten seasoned hematologists, the human specialists performed no better than random guessing when attempting to differentiate between real and AI-generated images.

That really surprised me. These are people who stare at blood cells all day, and even they couldn’t tell,” said Deltadahl.

As part of the project, the researchers are unveiling what they claim to be the largest publicly accessible dataset of peripheral blood smear images in the world, comprising over 500,000 images in total.

By making this resource open, we hope to empower researchers worldwide to build and test new AI models, democratize access to high-quality medical data, and ultimately contribute to better patient care,” said Deltadahl.

The researchers note that CytoDiffusion should not be viewed as a substitute for the expertise of qualified clinicians. Rather, it is intended to assist them by quickly identifying abnormal cases for further examination and managing more standard cases automatically.

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,” said Parashkev Nachev, Study Co-Senior Author and Professor, UCL.

Parashkev Nachev adds, “Our work suggests that generative AI will be central to this mission, transforming not only the fidelity of clinical support systems but their insight into the limits of their own knowledge. 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.”

The researchers say that further efforts are needed to improve the system's speed and to assess its performance across diverse patient demographics, ensuring fairness and accuracy.

Journal Reference:

Deltadahl, S., et al. (2025) Deep generative classification of blood cell morphology. Nature Machine Intelligence. DOI:10.1038/s42256-025-01122-7. 

Tell Us What You Think

Do you have a review, update or anything you would like to add to this news story?

Leave your feedback
Your comment type
Submit

Sign in to keep reading

We're committed to providing free access to quality science. By registering and providing insight into your preferences you're joining a community of over 1m science interested individuals and help us to provide you with insightful content whilst keeping our service free.

or

While we only use edited and approved content for Azthena answers, it may on occasions provide incorrect responses. Please confirm any data provided with the related suppliers or authors. We do not provide medical advice, if you search for medical information you must always consult a medical professional before acting on any information provided.

Your questions, but not your email details will be shared with OpenAI and retained for 30 days in accordance with their privacy principles.

Please do not ask questions that use sensitive or confidential information.

Read the full Terms & Conditions.