Ning Bi

I am a Postdoctoral Researcher at the University of Oxford, working with NDS and IBME on generative AI for healthcare, digital contrast CT, and scalable CT foundation models. My current research develops diffusion-based image translation methods that synthesize contrast-enhanced CT from non-contrast scans, with attention to spatial priors, uncertainty modelling, and clinical usability.

I completed my PhD in Computer Science at the University of Leeds, supervised by Prof. Alejandro F. Frangi and Dr. Zeike Taylor. My doctoral work focused on Bayesian deep learning for cardiac motion modelling, deformable registration, and joint segmentation-motion estimation in cardiac MRI.

I received my BSc in Computer Science with First Class Honours from ShanghaiTech University, where I worked on computer vision, object detection, and visual tracking as part of the SVIP Lab.

Email  /  CV  /  LinkedIn  /  Google Scholar  /  GitHub

My research sits at the intersection of medical image analysis, generative modelling, and foundation models for healthcare. I am especially interested in diffusion models, image-to-image translation, multimodal integration, uncertainty-aware learning, CT angiography, and cardiac MRI.

Developing and benchmarking generative models for contrast-enhanced CT synthesis from non-contrast CT. This work is connected to the NETZERO AICT consortium and broader efforts to build large-scale CT foundation models for scientific discovery and clinical translation.

Built probabilistic and deep learning methods for cardiac cine-MRI motion modelling, deformable registration, and concurrent segmentation-motion estimation.

Investigated model scaling, sparsity, and pruning strategies for large language models, with a focus on efficiency-performance tradeoffs in foundation model development.

* indicates equal contribution. A fuller and more current list is available on Google Scholar.

A multitask conditional diffusion framework for synthetic contrast-enhanced CT generation and aortic structure segmentation.

Consortium-scale CT analysis for cardiomegaly detection and prevalence assessment across an international imaging repository.

A recurrent variational framework for concurrent segmentation and motion estimation in cardiac cine-MRI sequences.

A hyperparameter-free deformable registration method using gradient surgery to balance registration accuracy and deformation smoothness.

Learning-based deformable registration for generating realistic cardiac MR sequences from a single frame.

Methods: diffusion models, Bayesian deep learning, image-to-image translation, deformable registration, multimodal data integration, uncertainty modelling, CT angiography, cardiac MRI.

Tools: Python, PyTorch, Hugging Face Transformers/Diffusers, MONAI, PyTorch Lightning, SciPy, scikit-learn, OpenCV, Pandas, Weights & Biases, Docker, Git, AWS, Microsoft Azure, Nibabel, ITK/SimpleITK, 3D Slicer.

Service and awards: reviewer for MICCAI, IEEE TMI, and IEEE TNNLS; Best Poster Award at the University of Leeds School of Computing Symposium; full PhD scholarship from the University of Leeds.