AI Tool Improves Blood Cell Analysis for Disease Detection

An AI tool that can analyze abnormalities in the shape and form of blood cells, and with greater accuracy and reliability than human experts, could change how conditions like leukemia are diagnosed.

Researchers have created a system called CytoDiffusion that uses generative AI – the same technology behind image generators such as DALL-E – to study the shape and structure of blood cells.

Unlike many AI models that simply recognize patterns, the researchers – from the University of Cambridge, University College London, and Queen Mary University of London – found that CytoDiffusion could accurately identify a wide range of normal blood cell appearances and spot unusual or rare cells that may indicate disease. Their results are published in the journal Nature Machine Intelligence.

Spotting subtle differences in blood cell size,shape,and appearance is key to diagnosing many blood disorders. But this task requires years of training,and even then,doctors can disagree on arduous cases.

We have many different types of blood cells with different properties and roles in our body. For example,white blood cells specialize in fighting infection. But knowing what an unusual or diseased blood cell looks like under a microscope is important for diagnosing many diseases.

Simon Deltadahl, Cambridge’s department of Applied Mathematics and Theoretical Physics, study’s first author

A typical blood ‘smear’ contains thousands of cells – far more than any human could analyze. “Humans can’t look at all the cells in a smear – it’s just not possible,” said Deltadahl. “Our model can automate this process, handle routine cases, and highlight anything unusual for a doctor to review.”

“When I was a junior haematology doctor, I faced many blood films to analyze after a day of work,” said co-senior author Dr Suthesh Sivapalaratnam from queen Mary University of London. “I became convinced AI would do a better job than me when analyzing them late at night.”

To develop CytoDiffusion, the researchers trained the system on over half a million images of blood smears collected at Addenbrooke’s Hospital in