Mathematical Tools for Ventricular Analysis using Cardiac MRI

Abstract

Non-invasive imaging techniques are now being used routinely for the analysis of cardiac function. The objective of this thesis was to develop mathematical modelling tools for the semi-automatic quantification of cardiac structure and function from magnetic resonance imaging (MRI). Two main problem areas were considered. Firstly, tools were developed to investigate the changes in cardiac function and myofibre structure during the progression of myocardial infarction, and the effect of using angiotensin-converting enzyme inhibitor (ACEI) as a treatment of myocardial infarction. Ex vivo diffusion tensor MRI (DTMRI) could then be compared with in vivo myocardial strain from MRI tissue tagging. These tools were applied to data from four healthy male Sprague Dawley rats, and eight with myocardial infarction induced by ligating the left anterior descending artery. Half of the infarcted rats were treated by ACEI. The results showed that myocyte structure as well as function were altered in myocardial infarction, altering the correlations between structure and function. A positive correlation between strain and fractional isotropy in the control group became negative in the infarct group and did not change with ACEI. Strain was positively correlated with the proportion of left-handed myofibres in the control group. This relationship was not significant in the infarct group but returned in the ACEI group. The results from these preliminary studies indicate that treatment with ACEI helps to restore normal myocardial structure-function relationships. Secondly, a modelling tool was developed for the efficient evaluation of right and left ventricular function in standard cine MRI imaging examinations. The biventricular modelling tool used a human biventricular deformable model, which was developed based on a porcine model, for customisation to cardiac MRI data. The customisation used an interactive guide point modelling technique which was modified to include a `predictor' step using a host mesh fitting algorithm, thereby obtaining a significant decrease in solution time. The tool was applied to cine MRI data of seventeen patients with various types of congenital heart disease. The results were compared against with those obtained from a current gold standard technique. The comparison showed generally good agreement between the two methods, in terms of both the reproducibility of global cardiac function measurements and the reproducibility between analysts. In conclusion, the tools developed in this thesis enabled novel examinations of cardiac structure and function in animal models and humans with cardiac disease.