Making MRI usable where the scanner is weak, the patient moves, and the report has to be checked.

My work builds agentic AI for medical imaging along three threads: motion-robust MRI reconstruction with implicit neural representations, ultra-low-field to high-field MRI synthesis via self-supervised diffusion, and verifiable radiology report generation with large language models. Before Imperial I worked on echocardiography analysis at Sun Yat-sen University.

I care about methods that are label-efficient, uncertainty-aware, and deployable, with the long-term aim of turning foundation-model-driven AI into tools radiologists can actually trust in a real workflow.