Research

Every heart is unique—and so is heart disease. Standard cardiac MRI scans provide static snapshots, but they don’t fully capture how an individual heart moves and functions over time. Our project focuses on developing personalized digital twins of the heart by fusing multi-view cardiac MRI to reconstruct 3D+time heart shape and motion. These models enable predictive simulations of disease progression and treatment response, offering a powerful tool for personalized diagnosis and care.

Medical imaging is essential for diagnosis, but AI development is often limited by data scarcery, privacy restrictions, and scanner differences. Generative AI offers a powerful solution—it can create realistic, synthetic images across imaging modalities. These synthetic scans expand training data, simulate rare cases, and harmonize images across hospitals, helping us build medical AI systems that are safer, fairer, and more robust.

Large language models (LLM) and vision-language models ((VLM)) are reshaping AI, but applying them in medicine comes with new challenges. Clinical data is multimodal, high-stakes, and often messy—making it hard to ensure consistent, reliable performance. Our research focuses on building trustworthy medical foundation models that combine images, text, and signals to support real-world clinical tasks. By improving transparency, robustness, and alignment with expert knowledge, we aim to develop medical AI that is not only powerful, but also safe, fair, and ready for deployment.

Medical AI models often perform well in controlled settings but struggle when applied to new hospitals, scanners, or patient populations. Our research focuses on building generalisable and robust medical AI that can handle real-world variability. By training and evaluating models across diverse datasets and domains, we aim to improve performance consistency, reduce bias, and ensure reliability under distribution shifts.