Dr Jinming Duan

Jinming Duan

Former position

Turing Fellow

Partner Institution

Bio

I am currently a Lecturer (Assistant Professor) within School of Computer Science at University of Birmingham (UoB) and a Visiting Researcher at Imperial College London (ICL). From 2018 to 2020, I was a Research Associate jointly within Department of Computing and Institute of Clinical Sciences at ICL. There, I have been working closely with Prof Daniel Rueckert and Prof Declan O’Regan, developing cutting-edge machine learning methods for cardiovascular imaging. From 2014 to 2018, I was studying my PhD degree in Computer Science at University of Nottingham under the supervision of Dr Li Bai, and my PhD thesis titled "Variational and Partial Differential Equation (PDE)-based Methods for Image Processing". In 2019, I developed an MRI reconstruction algorithm using deep learning which won 2nd place in the "fastMRI" competition organised by Facebook.

In 2016, I was awarded the "Chinese Government Award for Outstanding Self-financed Students Abroad" issued by Xiaoming Liu, the ambassador of Chinese Ministry of Education. In 2015, my master's thesis titled "A Study on Generalised Variational Level Set Methods and Their Fast Projection Algorithms" won the "Best Master Thesis Award" across the whole Shandong Province in China. In 2014, I was granted a full studentship from EPSRC to study my PhD at University of Nottingham. My work has appeared across 90+ peer-reviewed publications, including proceedings such as MICCAI, CVPR, ECCV, etc, and journals such as The Lancet, Nature Machine Intelligence, IEEE Transactions on Medical Imaging, IEEE Transactions on Image Processing, etc. I am now the UoB leading collaborator of the "smartHeart" programme funded by EPSRC UK.
 

Research interests

My research spans across multiple areas of artificial intelligence (AI) including computer vision, machine learning and medical imaging analysis. My research career began with the interaction of mathematical sciences and imaging, with a focus on model-based approaches (e.g. non-smooth and possibly non-convex variational approaches and non-linear partial differential equations) for image processing and inverse problems. Over time, my research interests have evolved into applications of these methods to more practical imaging problems, such as object detection, diffeomorphic registration, k-space magnetic resonance image (MRI) reconstruction, surface reconstruction from point clouds, etc.

I have extensively studied a variety of modern optimisation algorithms with advanced numerical discretization approaches, including finite difference, finite element and graph (nonlocal) methods. My current research has focused on: developing accelerated first order optimisation methods based on gradient descent for non-smooth or non-convex variational imaging models; leveraging domain knowledge (normally from model-based methods) for machinery algorithms such as deep neural networks to learn better from data; and applying these model-driven learning-based approaches to large-scale cardiac MRI datasets such as UK Biobank (containing scans from 100,000 subjects) and high-dimensional problems such as 4D cardiac image segmentation, shape modelling and motion tracking. My aim is to develop novel machine learning solutions for clinicians and cardiologists to understand the genetic and physiological mechanisms that underpin cardiovascular disease using cardiac imaging.