This thesis presents work towards a comprehensive set of methods to extract quantitative shape features and infarct information from cardiac magnetic resonance imaging data to enable classification of a heart disease state in terms of arrhythmia occurrence. The aim of the first part of this work is to process the images and convert the output to a computational domain usable for analysis / modelling. A comprehensive pipeline is constructed, to go from late gadolinium enhanced cardiac magnetic resonance images to numerical bi-ventricle meshes. The goal is to utilize the tools that are available open-source, in order to encourage the use by other researchers. A method of using mathematical currents to describe the meshes is employed to define the population mean shape and patient-specific shape characteristics. The resulting features are decomposed using linear and non-linear algorithms to extract the relevant information. The second part of this thesis concerns the application of machine learning algorithms to the set of subjects pre-processed with the pipeline. The main objective is to reduce the high variance imposed by the small size of available data. The unsupervised and supervised algorithms are used to search for patterns in the data and classify the myocardial infarction patients according to the presence of ventricular fibrillation during the first ST-elevation myocardial infarction to provide valuable insights into the pathophysiology. The single classifiers are combined using a variety of ensemble methods. Finally, a multiple kernel method which integrates multiple modalities is presented. This method takes advantage of the infarct information embedded in the late gadolinium enhanced cardiac magnetic resonance images. The classification of myocardial infarction patients used in the thesis is an example problem of using machine learning for clinical support. The techniques presented in this thesis can be used to analyze and classify any heart disease manifesting with ventricular remodelling and / or infarct development.