Imaging and genomics based multimodal foundational models for heart failure prognosis

Heart failure is a condition which can be severely life-limiting. It is a complex trait affecting over 30 million individuals worldwide and is influenced by environmental and genetic factors. Diagnosis and prognosis are determined using imaging and physiological measurements of the heart. While genomic technologies have led to the identification of a large number of genetic risk factors, they are not systematically used for diagnosis and prognosis. The project will develop novel machine learning methods, based on foundation models, to combine genomic information with cardiac imaging for higher accuracy of prognosis. A foundation model allows for the combination of multiple types of input and can generalise well across datasets. While these models have already shown promise, combined analyses of imaging and genomic data is a relatively unexplored area. A requirement of these methods is the availability of large sample sizes characterised with different measurements. The UK Biobank, which includes over 42,000 participants with both genomic and cardiac imaging data, is ideal for this approach. The main goal of this work is to improve our understanding of how imaging and genetic modalities can be usefully combined into predictive algorithms, a key technical challenge in enabling the next generation of prognostic tests. We will also train a model that, beyond this project, could be used to prototype future clinical deployments in identifying patients at high risk of heart failure.