Tailoring Deformable Models to Extract Meaningful Metrics and Landmarks from 3D Echocardiograms

Abstract

To combat heart disease, clinicians need accurate information about the patient's heart function. Ultrasound images of the heart can be used to provide this information, but analysis of the images is needed to get accurate assessments. This thesis is about new algorithms to measure heart functions related to the heart’s ability to pump blood, these are very important metrics for the heart’s health. In addition, this thesis studies how to automatically find important parts of the heart in the image, which saves time during a medical evaluation. The thesis consists of three papers. The first is about measuring the right ventricle’s ability to pump blood by accurately determining the ventricle’s borders in the image. The second paper is about finding the aorta and the mitral valve centre in certain ultrasound images. This information can be used to determine the standard views, a way of viewing the ultrasound image that gives the clinician much important information. The third paper is about making a single algorithm evaluate all the four chambers of the heart, the idea being that evaluating all at once saves time and that information about one chamber’s position can be used to properly find the other ones.