Characterising Shape Variation in the Human Right Ventricle Using Statistical Shape Analysis: Preliminary Outcomes and Potential for Predicting Hypertension in a Clinical Setting

Abstract

Variations in the shape of the human right ventricle (RV) have previously been shown to be predictive of heart function and long term prognosis in Pulmonary Hypertension (PH), a deadly disease characterised by high blood pressure in the pulmonary arteries. The extent to which ventricular shape is also affected by non-pathological features such as sex, body mass index (BMI) and age is explored in this thesis. If fundamental differences in the shape of a structurally normal RV exist, these might also impact the success of a predictive model. This thesis evaluates the extent to which non-pathological features affect the shape of the RV and determines the best ways, in terms of procedure and analysis, to adapt the model to consistently predict PH. It also identifies areas where the statistical shape analysis procedure is robust, and considers the extent to which specific, non-pathological, characteristics impact the diagnostic potential of the statistical shape model. Finally, recommendations are made on next steps in the development of a classification procedure for PH. The dataset was composed of clinically-obtained, cardiovascular magnetic resonance images (CMR) from two independent sources; The University of Pittsburgh Medical Center and Newcastle University. Shape change is assessed using a 3D statistical shape analysis technique, which topologically maps heart meshes through an harmonic mapping approach to create a unique shape function for each shape. Proper Orthogonal Decomposition (POD) was applied to the complete set of shape functions in order to determine and rank a set of shape features (i.e. modes and corresponding coefficients from the decomposition). MRI scanning protocol produced the most significant difference in shape; a shape mode associated with detail at the RV apex and ventricular length from apex to base strongly correlated with the MRI sequence used to record each subject. Qualitatively, a protocol which skipped slices produced a shorter RV with less detail at the apex. Decomposition of sex, age and BMI also derives unique RV shape descriptors which correspond to anatomically meaningful features. The shape features are shown to be able to predict presence of PH. The predictive model can be improved by including BMI as a factor, but these improvements are mainly concentrated in identification of healthy subjects.