Atlas-based Analysis of Biventricular Heart Shape and Motion in Congenital Heart Disease

Abstract

Quantifying heart shape and function changes is required to better understand the development of heart failure in congenital heart disease (CHD) and provide vital information for disease evaluation and therapy planning. It is important to analyse both left and right ventricles, and quantify 3D ventricular shape to characterise the morphological changes. In CHD, the ventricles remodel to very different geometries, making statistical comparison with reference populations difficult and the mass and volume measures used in clinical settings ignores a large part of the available information. Atlas-based analysis can provide detailed information on local shape variations and their relationships with disease processes. The aim of the thesis was to develop a novel method for the assessment of biventricular shape and function and characterisation of disease progression. A new 3D biventricular shape model representation was developed and validated on 60 participants with a variety of CHD pathologies. Using subdivision surface techniques, this new biventricular representation achieved better accuracy with fewer parameters than the previous finite element representation. It was flexible enough to characterise a wide range of CHD types. An iterative diffeomorphic algorithm for the registration of 3D biventricular models to 3D points was then developed. The algorithm was based on the decomposition of deformations, preventing the determinant of the Jacobian of the deformation falling below zero. Each transformation was guaranteed to be bijective by limiting the displacement of the coarse mesh within each iteration. The algorithm was tested on synthetic data, contours from 79 participants with a large variety of CHD types and contours from 4,989 UK Biobank participants. This technique was able to achieve good registration accuracy while generating diffeomorphic displacement fields. Biventricular models from the UK Biobank were used to build a biventricular atlas. Principal component analysis and logistic regression were used to extract shape variations due to cardiovascular risk factors present in the general population. Biventricular atlas features showed stronger associations between cardiovascular risk factors and heart shape than standard mass and volume measures. An atlas from 59 patients with tetralogy of Fallot was then built and shape changes within this cohort were quantified. Biventricular shape features relevant to pulmonary regurgitation were identified, enabling a continuous representation of shape and function changes with increasing severity of pulmonary regurgitation. Right ventricular dilatation and basal bulging, but also a left ventricular diminution were observed, suggesting a close interventricular relationship. TOF participants that have had a valve replacement were projected onto the UK Biobank space, giving current cardiac status and information on the outcome of the valve replacement. In conclusion, robust and accurate atlas-based methods have been developed and validated for the analysis of biventricular shape and function in CHD patients, as well as the general population.