MOSAIC

Currently, one in six people in the UK die from cancer, and many more live with its long-term effects. Artificial intelligence (AI) has the potential to drastically improve cancer care, but traditional AI models are narrow: they are built to do one specific task and require expensive, time-consuming manual data labelling to train.

Foundation models (FMs) solve this problem. They are large, powerful AI models trained on large amounts of data, making them highly adaptable "generalist" experts. Once trained, a single FM can then be efficiently tuned to perform dozens of different medical tasks. Furthermore, "multi-modal" FMs can read and connect different types of medical information at once, better understanding a patient's healthcare journey – for example the relationship between a patient's CT scan, their blood tests, and their past medical history.

Most current FMs are built using data from outside the UK, which means they may not suitably reflect the diverse patients seen within, and care delivered by, the NHS. The MOSAIC (Multi-modal Oncology Sovereign AI Collaboration) project will securely use NHS data to build a UK-specific cancer foundation model. We will test its ability to predict how well cancer treatments work, the likelihood of side effects, and long-term patient outcomes, ultimately aiming to personalise and improve cancer care.