Cardiovascular function assessment using computational blood flow modelling and machine learning

Student thesis: Doctoral ThesisDoctor of Philosophy

Abstract

Cardiovascular disease (CVD) is a leading cause of death and economic burden worldwide. Improving the early diagnosis and treatment of CVD would hugely benefit public health and boost the economy. Computational blood flow modelling and machine learning have been proven extremely beneficial in enhancing both diagnosis and treatment for CVD. This thesis consists of three original studies which aim at improving the assessment of cardiovascular function and diagnosis of CVD through either computational blood flow modelling or machine learning. The first study investigates the most prevalent clinical test for assessing endothelial dysfunction, flow mediated dilation (FMD), using one-dimensional (1-D) blood flow modelling coupled to a novel endothelial response function model. Although the FMD test is a widely used test for assessing endothelial function, the incremental value in predicting clinical outcomes is limited. The novel model developed in the first study enabled us to (i) confirm that the vasoconstriction occurring immediately after cuff deflation is a physical response entirely caused by a change in conduit artery transmural blood pressure and (ii) highlight important confounding factors that can affect the FMD test results such as arterial stiffness and blood pressure. These results help enhance the understanding of the limitations in the current FMD test and provide insights for further improvement. The second study assesses the accuracy of 1-D blood flow modelling to simulate pulse wave propagation across vascular abnormalities by using results from three-dimensional (3- D) blood flow modelling and measurements from an in vitro experiment as the ground truth. The 1-D approach has received notable attention in the last decades, and several studies have systematically assessed its accuracy in the larger systemic arteries under normal physiological conditions. However, less attention has been paid to systematically evaluate this modelling approach’s accuracy under pathological conditions (e.g. stenosis and aneurysm). The results in the second study show that the 1-D approach is able to capture the main features of the pressure and flow waveforms, the pressure drop across stenoses, and the energy dissipation across aneurysms in a similar manner to the 3-D and experimental approaches, with a relative error smaller than 7.5% for stenosis and aneurysm sizes up to 85% and 400%, respectively. The 1-D modelling approach for simulating haemodynamics in vasculatures with the stenoses and aneurysms provides a more efficient way of investigating these diseases numerically. The dataset created in the second study will support further development of 1-D numerical schemes in diseased arterial vasculatures. The third study aims to estimate pulse wave velocity (PWV), a well-established biomarker of vascular ageing (a risk factor for CVD), based on either the information extracted from a radial pressure wave or the whole waveform using machine learning approaches. The analysed results show that the PWV can be estimated from the radial pressure wave with a mean difference smaller than 0.2 m/s, and upper and lower limit of agreement smaller than 3.75 m/s and -3.34 m/s, respectively. Furthermore, the machine learning approach estimates PWV from an entire radial pressure wave is able to work with waveforms containing random noise with percentage errors increased by less than 2% when adding 20% noise to the waveform. Detecting early vascular ageing can reduce the morbidity and mortality of CVD. Using only a single peripheral pulse waveform, such as the radial pressure wave in the third study, can make CVD detection more accessible to a wider population. In summary, this thesis has assessed the accuracy of a currently available non-invasive clinical test, and provided tools for possible new non-invasive clinical assessments of CVD that could help early detection and monitoring of the disease.
Date of Award1 Apr 2021
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorJordi Alastruey (Supervisor) & Philip Chowienczyk (Supervisor)

Cite this

'