Rechercher des projets européens

Variational Basis Learning for Statistical Motion Atlases: Application to Quantitative Dynamic Cardiac Imaging (BALMORAL)
Date du début: 1 juil. 2014, Date de fin: 30 juin 2016 PROJET  TERMINÉ 

Pulmonary Arterial Hypertension (PAH) is a severe progressive disorder characterised by a vasculopathy of the small pulmonary arteries to the lung. Failure of the right ventricle (RV) to adapt to elevated resistance to blood flow results in death, usually within 3 years for untreated patients with PAH. Image-based global measures can only reflect the overall performance of the RV; however, there is good evidence that PAH can be identified by localised motion abnormalities in the RV and the interventricular septum (IVS), without the need for invasive and expensive right heart catheterisation.In this proposal, we are interested in assessment of the diagnostic value of the motion abnormalities in RV and IVS, relevant to PAH. Given two groups of PAH patients and healthy controls, an important distinguishing feature of our computational framework with the existing literature is that it will allow multiscale evaluations all at the same time: L1) At the population level; a statistical motion atlas describing the “average” pattern of the heart motion over the population will be constructed. Two atlases will be made for PAH patients and healthy control subjects; L2) at the patient level; for any subject a probability value of being a patient with PAH will be measured to describe the severity of the disease; L3) at the myocardium level; localised and expert interpretable abnormality map over the heart will be measured for a given patient. From two populations of patients with PAH and normal controls, we aim to learn a set of optimal basis functions that are both discriminative at the patient level, and sparsely fitted to the pathological areas.The proposed method is a novel full Bayesian probabilistic framework, which learns the sparseness and the number of the basis function from the data by maximising the model evidence using variational Bayes.