DeepGeM: Deep Generative Modelling for Understanding Cardiac Anatomy and Function from Large-Scale Imaging Datasets
Lead Research Organisation:
Imperial College London
Department Name: Brain Sciences
Abstract
Cardiovascular disease is the leading cause of death globally. Structural and functional parameters derived from cardiac imaging data provide important indicators for diagnosing and managing cardiovascular disease. However, most of these parameters, such as the chamber volumes of the heart, only focus on the global description of its anatomy. Detailed analysis of the 3D-t (or 4D) cardiac anatomical data is often ignored in diagnosis due to the complexity of high-dimensional data modelling. This hinders our understanding of cardiac anatomy and function, as well as their longitudinal evolution in ageing and disease progression.
The ambition of this project is to develop a fundamentally new approach for analysing 4D cardiac imaging data to understand both the anatomy and function. We will develop novel generative machine learning methods for modelling the variations of cardiac anatomy among different people and across time. The generative machine learning model will be trained using large-scale cardiac imaging datasets. By combining both cardiac imaging data and non-imaging clinical data, the model will learn not only the spatio-temporal variations of cardiac anatomy but also how clinical factors influence the anatomy and function. We will demonstrate the clinical usefulness of the model in two tasks, namely to understand the cardiac anatomy in different groups of people and to predict the longitudinal change of cardiac anatomy and function. The ability to predict the cardiac anatomy and function in the future will potentially provide cardiologists with new tools for managing cardiac patients in personalised healthcare.
The ambition of this project is to develop a fundamentally new approach for analysing 4D cardiac imaging data to understand both the anatomy and function. We will develop novel generative machine learning methods for modelling the variations of cardiac anatomy among different people and across time. The generative machine learning model will be trained using large-scale cardiac imaging datasets. By combining both cardiac imaging data and non-imaging clinical data, the model will learn not only the spatio-temporal variations of cardiac anatomy but also how clinical factors influence the anatomy and function. We will demonstrate the clinical usefulness of the model in two tasks, namely to understand the cardiac anatomy in different groups of people and to predict the longitudinal change of cardiac anatomy and function. The ability to predict the cardiac anatomy and function in the future will potentially provide cardiologists with new tools for managing cardiac patients in personalised healthcare.
Organisations
Publications
Chen C
(2022)
Enhancing MR image segmentation with realistic adversarial data augmentation
in Medical Image Analysis
Curran L
(2023)
Genotype-Phenotype Taxonomy of Hypertrophic Cardiomyopathy.
in Circulation. Genomic and precision medicine
Meng Q
(2023)
DeepMesh: Mesh-based Cardiac Motion Tracking using Deep Learning
in IEEE Transactions on Medical Imaging
Meng Q
(2022)
MulViMotion: Shape-Aware 3D Myocardial Motion Tracking From Multi-View Cardiac MRI.
in IEEE transactions on medical imaging
Ouyang C
(2023)
Causality-Inspired Single-Source Domain Generalization for Medical Image Segmentation
in IEEE Transactions on Medical Imaging