UCL researchers have developed a pioneering virtual trial platform that could dramatically improve personalised treatment recommendations for patients suffering from ischaemic stroke; the most common type of stroke. By running more than 100 million biologically realistic simulations, the team found that traditional, simplified models used to guide stroke care may miss crucial individual differences in brain structure and biology.
Ischaemic stroke occurs when a blood vessel to the brain becomes blocked, and timely treatment is critical. Current clinical practice relies on results from large randomised controlled trials, which average treatment effects across broad groups of patients. But the human brain is highly variable, and strokes themselves differ widely in size, location, and biological impact. This complexity means the “average” treatment effect may not always be the best guide for individual patients.
To address this challenge, the research team created an advanced semi-synthetic virtual trial framework, combining large-scale maps of brain connectivity and function, genetic expression and receptor distribution data and more than 4,000 real acute stroke lesion maps aligned to a common brain template.
This approach allowed the researchers to simulate realistic patterns of disability after stroke and the likely biological responsiveness to treatment, producing a rich and detailed “ground truth” against which predictive models could be tested.
Inside each virtual trial, the team compared simple prediction models, similar to those commonly used in clinical trials, with more complex models that incorporated high-resolution lesion information and biological data. Even when the data fed into the complex models included noise and confounding, conditions that mimic the realities of clinical practice, the complex models consistently provided more accurate personalised treatment predictions.
This breakthrough suggests that future stroke care could move beyond one-size-fits-all recommendations toward approaches that tailor treatments to the unique anatomy and biology of each patient’s brain.
By harnessing large-scale, empirically grounded virtual trials, something that would be impossible to conduct with real patients, the research opens a new avenue for developing precision medicine tools in stroke care.
Lead author, Dr Dominic Giles (UCL Queen Square Institute of Neurology) said: “Just as each of us is unique, so too are the ways diseases such as stroke individually affect us. The characteristics that shape this individuality are highly intricate and difficult to capture. Yet treatment decisions often rely on crude characteristics such as symptom duration or stroke overall size. By applying advanced AI methods to finely detailed stroke imaging data, we show that these subtle, individual differences, which may determine whether a patient responds to treatment, can be learned, establishing a framework for developing more personalised treatment strategies in the future.”
Senior author, Professor Parashkev Nachev (UCL Queen Square Institute of Neurology) said: “Healthcare AI is widely seen as luxury technology, used to automate tasks human beings can do as well or better. But its real value lies in enabling us to describe patients objectively with the richness the complexity of their individuality demands. It allows us to predict with high fidelity not just the factual but also the counterfactual; what could have happened had an individual patient received an alternative treatment. Here we show that this approach may prove to be essential to achieving truly personalised care in stroke.”