Jiang, Yue; (2025) AI for Maximising Information from Cardiac MRI: Application to the Aorta and Congenital Heart Disease. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Jiang_10219389_Thesis.pdf Download (6MB) | Preview

Abstract

Magnetic resonance imaging (MRI) is a powerful tool for cardiovascular assessment, but high-resolution 3D scans require long acquisition times and high costs, limiting their routine use in clinics. Meanwhile, routine clinical MRI is more accessible but typically low-resolution, restricting accurate 3D assessment. This thesis investigates how artificial intelligence (AI), particularly deep learning, can recover high-resolution detail from rapidly acquired scans and maximise the 3D information extracted from routine cardiac MRI, with a focus on the aorta and congenital heart disease through the following three studies: First, an automated deep learning pipeline was developed to localise and segment the thoracic aorta from routine MRI, enabling fully automated measurements of 3D geometric features. Validation on both prospectively acquired data and the UK Biobank showed good agreement with manual segmentation made directly on high-resolution 3D MRI and confirmed suitability by clinicians. This framework shows strong potential for large-scale screening of aortic aneurysm. Second, the pipeline from the first project was extended with an additional deep learning module to measure aortic pulse wave velocity (PWV), a well-established marker of aortic stiffness, directly from routine clinical MRI. The fully automated method, requiring no user interaction, was validated on the UK Biobank, demonstrating its potential for large-scale, retrospective assessment of aortic stiffness. Finally, a super-resolution model was developed to enhance rapidly acquired low-resolution whole-heart MRI into high-resolution whole-heart images. Validation on a prospective dataset showed that the model could recover images acquired in ~9 seconds to a quality comparable with conventional high-resolution whole-heart scans acquired in ~6 minutes. This approach offers a potential route to reliable single-breath-hold whole-heart imaging in clinical practice. Together, these studies demonstrate how deep learning can increase the clinical value of routine cardiac MRI, supporting broader adoption in both research and practice.

Download activity - last month

Export as CSVJSONXML

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item