A new AI system called CytoDiffusion, developed by researchers from the University of Cambridge, University College London, and Queen Mary University of London, is showing promise in identifying blood cell abnormalities that may be missed by human doctors. The system uses generative AI—similar to the technology behind image generators like DALL-E—to analyze the shape, size, and structure of blood cells in microscopic images. Unlike traditional AI models that are trained to classify cells into predefined categories, CytoDiffusion learns the full range of normal cell appearances and can detect subtle or rare abnormalities that could signal diseases such as leukemia. The model was trained on over half a million blood smear images collected from Addenbrooke’s Hospital in Cambridge, making it the largest dataset of its kind. It includes both common and rare cell types, as well as variations caused by different microscopes, staining techniques, and hospital settings. The researchers found that CytoDiffusion outperformed existing systems in detecting abnormal cells linked to blood disorders, even with limited training data. It also demonstrated a key advantage over human experts: the ability to quantify its own uncertainty. When the model is unsure, it acknowledges it—reducing the risk of false confidence, a problem that can affect even experienced clinicians. In a test where 10 expert haematologists were asked to distinguish real blood cell images from AI-generated ones, the human experts performed no better than chance. This shows that the AI can produce highly realistic, indistinguishable images, a sign of its advanced learning capability. The team has released the entire dataset publicly, aiming to support global research and help democratize access to high-quality medical data. The goal is to accelerate the development of better diagnostic tools and improve patient care. While CytoDiffusion is not intended to replace doctors, it is designed to assist them by automating routine analysis and flagging unusual cases for expert review. The researchers emphasize that the true value of AI in healthcare lies not in replacing human expertise, but in enhancing it—by providing deeper insights and a clearer understanding of the model’s own limitations. The work, published in Nature Machine Intelligence, highlights the potential of generative AI to bring a new level of precision and self-awareness to medical diagnostics. The team is now working to improve the system’s speed and test its performance across diverse populations to ensure fairness and broad applicability. The research was supported by the Trinity Challenge, Wellcome, the British Heart Foundation, Cambridge University Hospitals NHS Foundation Trust, Barts Health NHS Trust, the NIHR Cambridge and UCLH Biomedical Research Centres, and NHS Blood and Transplant. It was conducted by the Imaging working group of the BloodCounts! consortium, which aims to transform blood diagnostics using artificial intelligence.