Abstract:
Heart failure (HF) is the leading cause of mortality in the western world. There is an urgent
need for more personalised and improved diagnosis, prognosis, and clinical management of
heart disease. Current diagnosis techniques of HF are limited in terms of their accuracy and
the specificity of the information they provide. Research has shown the importance of
regional myocardial function measures such as strain and torsion in early identification of
dysfunction. Recent advances in magnetic resonance imaging (MRI) have seen the
development of the Displacement Encoding with Stimulated Echoes (DENSE) technique,
allowing high resolution measurements of regional intramyocardial motion by encoding
tissue displacements in the images. This has motivated the need for fast and accurate image
analysis tools that can estimate strain measures and be implemented in the clinical setting.
This thesis presents a novel framework for the estimation of 3D strain fields from cine
DENSE images. 3D kinematic free-form deformation (FFD) modelling methods were used
to generate full 3D cylindrical strain fields for a mid-ventricular ring of myocardial tissue to
provide information about regional intramyocardial tissue function. The FFD framework was
developed and validated with synthetic cine DENSE images generated using a computational
phantom subject to cardiac-like deformations. For this, a new deformation model was
presented, expressed in cylindrical polar coordinates, that involved non-zero values for all 6
components of strain. Modelling and imaging recommendations were made through in-depth
error analysis. It was found that partial volume effects had the most significant effect,
particularly on radial strain estimates. Finally, the application to in vivo cine DENSE images
acquired from a cardiac patient was shown. The resulting FFD framework strains showed
good agreement with strains derived from the existing 2D image analysis tool,
DENSEanalysis. Further investigation is required to improve the accuracy of image
acquisition and the FFD framework to ensure the generation of reliable patient-specific
models that are ready for use in the clinical environment.