Towards 10-minute Magnetic Resonance Scanning in Children - Developing Accelerated Imaging Using Machine Learning

Project ID: 2228cd1434 (You will need this ID for your application)

UCL Lead department: Institute of Cardiovascular Science

Project Summary:

Magnetic Resonance Imaging (MRI) plays a vital role in diagnosis, management and monitoring of many childhood diseases. MRI enables true three-dimensional volumes to be collected for comprehensive assessment of anatomy. However, acquisitions are time-consuming, resulting in long overall scan times and/or artefacts due to motion. One way of overcoming these problems is to speed up MRI scanning by collecting less data, but this causes artefacts in the images. Mathematical algorithms to reconstruct clinically useful images exist, but are too time-consuming to use in a hospital setting.

This project will develop Machine Learning (ML) techniques to enable rapid 3D MRI in childhood diseases, reducing total scan times from ~1hour to ~10minutes. The clinical applications will primarily cover cardiac and abdominal imaging, where motion is particularly problematic.

ML will be developed to perform denoising/artefact suppression of vastly undersampled 3D-MRI data. Synthetic paired training data will be created from our large curated, gold-standard k-space datasets, to simulate the artefacts caused by fixed-trajectory undersampling. 3D U-Nets will be optimised to reconstruct good image quality, in clinically-useable times. Building on this, we will focus on joint optimisation of the sampling pattern and reconstruction, to find the ‘best’ acquisition pattern to enable higher resolution 3D imaging. Finally, we will investigate the use of ML to perform motion correction by simulating motion in the training data, to further reduce the need for patent cooperation and enable free-breathing scans.

The student will be integrated into our enthusiastic, multi-disciplinary team, who have extensive experience in fast MRI techniques, motion correction and ML. The student will learn how MRI works, current state-of-the-art reconstruction strategies, novel ML techniques and motion correction strategies. The student will build on our open-source ML framework for MRI reconstruction, and the existing pipelines to integrate the resulting models into the scanner environment for prospective validation.