Data-driven Representation Learning for Equitable, Personalised Care in Multiple Sclerosis: DREAMS

Most hospital data for chronic conditions like multiple sclerosis (MS)—from narrative notes to medical images—is recorded in formats that defy systematic analysis, leaving it underused. This Fellowship will pioneer foundational machine learning algorithms to unlock this data to enable personalising care. Through a unique international collaboration across 16 centres and four countries, I will employ decentralised machine learning to build robust, generalisable models. I will deliver on three objectives: (1) ultrarobust medical image (MRI) algorithms able to quantify brain and spinal cord damage even from imperfect, multi-site clinical scans; (2) advanced natural language models that distil complex clinical notes into data-derived measures of disease evolution; and (3) novel multimodal algorithms that integrate medical images and text to predict disability progression, stratify patients into subtypes, and assist with clinical image reporting. This Fellowship will pioneer a scientific toolbox for leveraging unstructured real-world data with potential to be applied across neurological diseases. Ultimately, it will enable earlier and more personalised clinical decisions to delay disability, maintain employment, and reinvest savings into healthcare systems.