Artificial intelligence clinical decision support systems for cardiovascular magnetic resonance imaging in aortic valve disease
File(s)
Author(s)
Vimalesvaran, Kavitha
Type
Thesis
Abstract
Cardiovascular magnetic resonance (CMR) imaging is increasingly used to assess cardiac structure, function, and myocardial disease, including the myocardial consequences of aortic valve disease. However, CMR workflows remain complex, time-intensive, and highly dependent on operator expertise, contributing to variability, inefficiency, and potential diagnostic uncertainty. This thesis explores the application of artificial intelligence (AI) to address specific limitations within the CMR workflow, with a focus on aortic valve disease.
In Chapter 3, I report findings from a cross-sectional survey of UK radiographers and trainee clinicians performing CMR. The results demonstrate variability in protocol confidence, moderate dissatisfaction with workflow efficiency, and strong support for AI-based clinical decision support, helping to define clinically relevant targets for AI intervention.
Chapter 4 introduces a novel imaging biomarker for aortic stenosis: the ratio of blood signal intensity between the ascending aorta and left ventricle (Ao:LV), derived from standard three-chamber cine bSSFP images. This biomarker was retrospectively validated in a multicentre cohort of 314 patients and showed consistent association with echocardiographic measures of disease severity.
In Chapter 5, I develop a deep learning framework to detect abnormal aortic valve pathology directly from three-chamber cine images. A landmark- and curve-based approach was used to characterise valve-related signal changes, prioritising interpretability and alignment with known haemodynamic features.
Chapter 6 focuses on automating ratiometric biomarkers, including Ao:LV and Ao:LA, using task-specific deep learning models. Systematic evaluation demonstrated that different spatial sampling strategies optimise distinct diagnostic tasks, highlighting the importance of method selection based on clinical intent.
Overall, this thesis presents a clinically grounded investigation of AI applications across the CMR workflow, emphasising interpretability, feasibility, and relevance to real-world practice.
In Chapter 3, I report findings from a cross-sectional survey of UK radiographers and trainee clinicians performing CMR. The results demonstrate variability in protocol confidence, moderate dissatisfaction with workflow efficiency, and strong support for AI-based clinical decision support, helping to define clinically relevant targets for AI intervention.
Chapter 4 introduces a novel imaging biomarker for aortic stenosis: the ratio of blood signal intensity between the ascending aorta and left ventricle (Ao:LV), derived from standard three-chamber cine bSSFP images. This biomarker was retrospectively validated in a multicentre cohort of 314 patients and showed consistent association with echocardiographic measures of disease severity.
In Chapter 5, I develop a deep learning framework to detect abnormal aortic valve pathology directly from three-chamber cine images. A landmark- and curve-based approach was used to characterise valve-related signal changes, prioritising interpretability and alignment with known haemodynamic features.
Chapter 6 focuses on automating ratiometric biomarkers, including Ao:LV and Ao:LA, using task-specific deep learning models. Systematic evaluation demonstrated that different spatial sampling strategies optimise distinct diagnostic tasks, highlighting the importance of method selection based on clinical intent.
Overall, this thesis presents a clinically grounded investigation of AI applications across the CMR workflow, emphasising interpretability, feasibility, and relevance to real-world practice.
Version
Open Access
Date Issued
2025-05-20
Date Awarded
2026-03-01
Copyright Statement
Attribution-NonCommercial 4.0 International Licence (CC BY-NC)
License URL
Advisor
Bharath, Anil
Cole, Graham
Publisher Department
Department of Computing
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)