Mehrdad Yaghoobi, m.yaghoobi-vaighan@ed.ac.uk
Lucy Kershaw , Lucy.Kershaw@ed.ac.uk
Project Description
Cardiac Magnetic Resonance Imaging (Cardiac MRI) has been used as a non-invasive method for medical diagnostics. It aims to incorporate the heart movements and deformations and reconstruct 3D, or 3D+time, images. However, there are challenges in imaging with fewer K-space samples, for fast imaging, and making the process self-driving to facilitate the widespread use of this approach.
For cardiac imaging this is compounded by the need for breath-holding and ECG gating to compensate for respiratory and cardiac motion. Various methods have been proposed to accelerate the acquisition time, including multi-element and compressed sensing frameworks, but cardiac MRI is still less popular than conventional MRI, partly due to the complicated scanning protocols and image prescription.
This project wants to use recent deep learning methodologies in self-supervised learning to avoid labelling biases (due to ambiguities in labels), and to use a larger corpus of unlabelled data, through an underlying discovered structure, called representation space. Specifically, we aim to improve the robustness of image reconstructions and analyses, and make the solutions scanner-agnostic, eventually implementing this in an open reconstruction framework that allows user algorithms to reconstruct images on the scanner at run-time rather than requiring off-line reconstruction.
The PhD researcher will learn about the physics and clinical aspects of Cardiac MRI, signal acquisition and processing, and machine learning techniques during this project. The student will have the opportunity to engage with cardiologists, radiologists, and radiographers to understand their requirements for patient comfort, straightforward acquisition, and image quality.
