Atlas Based Analysis of Heart Shape and Motion in Cardiovascular Disease

Abstract

Atlas-based analyses of patients and healthy volunteers have recently been explored in several different medical areas. Large imaging databases have been established, which enable the construction of probabilistic shape atlases for specific organs or diseases. The Cardiac Atlas Project (CAP) is a world-wide web-accessible resource, comprising a population atlas of asymptomatic and pathological hearts. Large numbers of cardiac MRI data have been contributed from several studies including Defibrillators to Reduce Risk by Magnetic Resonance Imaging Evaluation (DETERMINE) study, comprising patients with myocardial infarction, and the Multi Ethnic Study of Atherosclerosis (MESA) study, comprising asymptomatic volunteers. Finite element models have been customized to facilitate statistical shape analysis. The aim of the study was to quantify the cardiac shape difference between asymptomatic and myocardial infarction (MI) cases at the population level. Several shape feature extraction techniques were explored. We started with an unsupervised feature extraction techniquesprincipal component analysis (Chapter 2). In Chapter 3, two supervised feature extraction techniques－ information maximizing component analysis and linear discriminant analysis – were investigated to extract the most discriminatory global shape changes associated with remodelling after MI. In Chapter 4, we applied partial least square regression to derive shape features which are relevant to traditional clinical indices. We developed a novel method for deriving orthogonal shape decompositions directly from any set of clinical indices including left ventricle (LV) size, sphericity, wall thickness, ejection fraction, apical conicity and longitudinal shortening. In Chapter 5, we studied 13 traditional LV remodelling indices based on a casecontrol study on 408 MI cases from the DETERMINE study and 1991 asymptomatic subjects from the MESA study. In Chapter 6, we applied the shape analysis at the LV regional level and studied regional LV shape changes due to myocardial infarction. Results showed that our framework quantitatively characterized remodelling features associated with myocardial infarction. These features have the potential to enable precise quantification of the amount of remodelling present in a patient, and can be used to explore the effect of treatments designed to reverse remodelling effects.