UCL Discovery
UCL home » Library Services » Electronic resources » UCL Discovery

Automated deep phenotyping of the cardiovascular system using magnetic resonance imaging

Bhuva, Anish Naresh; (2020) Automated deep phenotyping of the cardiovascular system using magnetic resonance imaging. Doctoral thesis (Ph.D), UCL (University College London). Green open access

[thumbnail of PhD_Thesis_AnishBhuva_27112019_cleaned.pdf]

Download (26MB) | Preview


Across a lifetime, the cardiovascular system must adapt to a great range of demands from the body. The individual changes in the cardiovascular system that occur in response to loading conditions are influenced by genetic susceptibility, and the pattern and extent of these changes have prognostic value. Brachial blood pressure (BP) and left ventricular ejection fraction (LVEF) are important biomarkers that capture this response, and their measurements are made at high resolution. Relatively, clinical analysis is crude, and may result in lost information and the introduction of noise. Digital information storage enables efficient extraction of information from a dataset, and this strategy may provide more precise and deeper measures to breakdown current phenotypes into their component parts. The aim of this thesis was to develop automated analysis of cardiovascular magnetic resonance (CMR) imaging for more detailed phenotyping, and apply these techniques for new biological insights into the cardiovascular response to different loading conditions. I therefore tested the feasibility and clinical utility of computational approaches for image and waveform analysis, recruiting and acquiring additional patient cohorts where necessary, and then applied these approaches prospectively to participants before and after six-months of exercise training for a first-time marathon. First, a multi-centre, multi-vendor, multi-field strength, multi-disease CMR resource of 110 patients undergoing repeat imaging in a short time-frame was assembled. The resource was used to assess whether automated analysis of LV structure and function is feasible on real-world data, and if it can improve upon human precision. This showed that clinicians can be confident in detecting a 9% change in EF or a 20g change in LV mass. This will be difficult to improve by clinicians because the greatest source of human error was attributable to the observer rather than modifiable factors. Having understood these errors, a convolutional neural network was trained on separate multi-centre data for automated analysis and was successfully generalizable to the real-world CMR data. Precision was similar to human analysis, and performance was 186 times faster. This real-world benchmarking resource has been made freely available (thevolumesresource.com). Precise automated segmentations were then used as a platform to delve further into the LV phenotype. Global LVEFs measured from CMR imaging in 116 patients with severe aortic stenosis were broken down into ~10 million regional measurements of structure and function, represented by computational three-dimensional LV models for each individual. A cardiac atlas approach was used to compile, label, segment and represent these data. Models were compared with healthy matched controls, and co-registered with follow-up one year after aortic valve replacement (AVR). This showed that there is a tendency to asymmetric septal hypertrophy in all patients with severe aortic stenosis (AS), rather than a characteristic specific to predisposed patients. This response to AS was more unfavourable in males than females (associated with higher NT-proBNP, and lower blood pressure), but was more modifiable with AVR. This was not detected using conventional analysis. Because cardiac function is coupled with the vasculature, a novel integrated assessment of the cardiovascular system was developed. Wave intensity theory was used to combine central blood pressure and CMR aortic blood flow-velocity waveforms to represent the interaction of the heart with the vessels in terms of traveling energy waves. This was performed and then validated in 206 individuals (the largest cohort to date), demonstrating inefficient ventriculo-arterial coupling in female sex and healthy ageing. CMR imaging was performed in 236 individuals before training for a first-time marathon and 138 individuals were followed-up after marathon completion. After training, systolic/diastolic blood pressure reduced by 4/3mmHg, descending aortic stiffness decreased by 16%, and ventriculo-arterial coupling improved by 14%. LV mass increased slightly, with a tendency to more symmetrical hypertrophy. The reduction in aortic stiffness was equivalent to a 4-year reduction in estimated biological aortic age, and the benefit was greater in older, male, and slower individuals. In conclusion, this thesis demonstrates that automating analysis of clinical cardiovascular phenotypes is precise with significant time-saving. Complex data that is usually discarded can be used efficiently to identify new biology. Deeper phenotypes developed in this work inform risk reduction behaviour in healthy individuals, and demonstrably deliver a more sensitive marker of LV remodelling, potentially enhancing risk prediction in severe aortic stenosis.

Type: Thesis (Doctoral)
Qualification: Ph.D
Title: Automated deep phenotyping of the cardiovascular system using magnetic resonance imaging
Event: UCL
Open access status: An open access version is available from UCL Discovery
Language: English
Additional information: Copyright © The Author 2020. Original content in this thesis is licensed under the terms of the Creative Commons Attribution 4.0 International (CC BY 4.0) Licence (https://creativecommons.org/licenses/by/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request.
UCL classification: UCL
UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences
UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Population Health Sciences > Institute of Health Informatics
URI: https://discovery.ucl.ac.uk/id/eprint/10092907
Downloads since deposit
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item