What is BloodCounts!?
BloodCounts! is the first healthcare project hosted by the Cambridge Centre for Advanced Research and Education in Singapore, known as CARES, in the centre's 13-year history. CARES is the University of Cambridge's first research centre outside the UK. The project is led by Professor Weisi Lin of Nanyang Technological University and Professor Michael Roberts of the University of Cambridge, and brings together mathematicians, clinicians, and software engineers from Singapore and the UK. Its hospital partners are the National University Health System (NUHS) and SingHealth, joining as the first Singapore partners in a consortium that also includes centres in Belgium, The Gambia, Ghana, India, and the Netherlands. The project is supported by Singapore's National Research Foundation under its AI-for-Science initiative, as one of eight projects chosen to speed up discovery with AI. A launch event is planned for Thursday, October 29.
Why a humble blood count holds hidden information
Healthcare systems perform about 10 million complete blood count tests every day. Each test measures the cells in a blood sample, and doctors already use it to check for anaemia, infection, and many other conditions. But according to the project team, much of the information in those tests is not fully used, even though it contains valuable medical signals. BloodCounts! will use CBC data to build frontier AI models that discover subtle alterations in blood cells across a range of diseases. The AI is not looking for one obvious red flag. It is learning to notice small, complex patterns that a simple threshold on a lab report might miss, supporting earlier diagnosis, better prognosis, and more informed clinical decisions. This quiet, data-first approach is part of a wider shift in AI health research, where tools that read signals the body already gives us can open new options — much like the brain-computer interface work that gave unresponsive ICU patients a way to communicate, as covered in our article on Columbia's brain-computer interface study (https://seriouslyitsai.blogspot.com/2026/10/brain-computer-interface-gives.html).
How the AI will be built — and why diversity matters
NUHS will provide robust, anonymised CBC and clinical data, plus input from experts in haematology, laboratory medicine, oncology, neurology, and biomedical informatics, so Singapore's population is represented in a large, diverse international dataset. SingHealth brings trusted clinical data infrastructure and translational research experience spanning cancer, neuroscience, clinical informatics, and artificial intelligence. Professor Roberts has said that deep partnership between mathematicians, clinicians, and software engineers is what allows the team to build models anchored in rigorous mathematics, informed by real clinical practice, and available openly to researchers globally. Working across hospitals in Europe, Africa, and Asia gives the project the diversity of patients needed to build equitable AI models — crucial in medicine, where a tool that works well for one population but poorly for another can do real harm. The same care around evidence is visible in other AI health work, such as the $400M+ agreement between Astellas and AI Proteins to design miniprotein drugs from scratch (https://seriouslyitsai.blogspot.com/2026/10/astellas-and-ai-proteins-sign-400m-deal.html).
What this could mean for patients and doctors
For patients, the most attractive part is what BloodCounts! does not ask for: no new test, no extra sample. The value lies in using information already generated in routine care. If the models work as hoped, a test taken for one reason could also flag a risk that deserves a closer look, or help a doctor stratify risk more precisely and personalise care. For doctors, the tools are meant to support decision-making, not replace it — predictive tools that sit alongside clinical judgment. AI in healthcare works best when it augments a clinician, giving them a sharper lens on data they already trust. Similar hopes surround AI-designed medicines, and our report on an AI-designed drug reaching human trials (https://seriouslyitsai.blogspot.com/2026/10/an-ai-designed-drug-just-reached-human.html) shows how long the road from model to clinic can be, even when the early science is exciting.
What to watch next
After the October 29 launch, the real test begins: turning CBC data into validated models, proving they work across different hospitals and populations, and showing they genuinely improve outcomes rather than just producing interesting patterns. Watch for three things: whether the models are validated on data from hospitals that did not train them, whether the team publishes its methods openly so others can check them, and whether any tool is tested in real clinical workflows before anyone claims it is ready for routine use. BloodCounts! is a research project, not a product launch, and it would be wrong to promise patients a new test tomorrow. But as a direction of travel, it is hard to argue with: use the data we already collect, build the models carefully and openly, and let earlier, fairer detection be the reward.
BloodCounts! is a reminder that the most useful AI in healthcare may not be the flashiest. It may be the one that quietly reads a test you already take and helps your doctor see something sooner. If a routine blood test could flag a risk months earlier, with no extra needle and no extra cost to you, would you want your doctor to use it? Tell us in the comments.
Comments
Post a Comment