Three-dimensional model-based analysis of vascular and cardiac images

Abstract

This thesis is concerned with the geometrical modeling of organs to perform medical image analysis tasks. The thesis is divided in two main parts devoted to model linear vessel segments and the left ventricle of the heart, respectively. Chapters 2 to 4 present different aspects of a model-based technique for semi-automated quantification of linear vessel segments from 3-D Magnetic Resonance Angiography (MRA). Chapter 2 is concerned with a multiscale filter for the enhancement of vessels in 2-D and 3-D angiograms. Chapter 3 applies the filter developed in Chapter 2 to determine the central vessel axis in 3-D MRA images. This procedure is initialized using an efficient user interaction technique that naturally incorporates the knowledge of the operator about the vessel of interest. Also in this chapter, a linear vessel model is used to recover the position of the vessel wall in order to carry out an accurate quantitative analysis of vascular morphology. Prior knowledge is provided in two main forms: a cylindrical model introduces a shape prior while prior knowledge on the image acquisition (type of MRA technique) is used to define an appropriate vessel boundary criterion. In Chapter 4 an extensive in vitro and in vivo evaluation of the algorithm introduced in Chapter 3 is described. Chapters 5 to 7 change the focus to 3D cardiac image analysis from Magnetic Resonance Imaging. Chapter 5 presents an extensive survey, a categorization and a critical review of the field of cardiac modeling. Chapter 6 and Chapter 7 present successive refinements of a method for building statistical models of shape variability with particular emphasis on cardiac modeling. The method is based on an elastic registration method using hierarchical free-form deformations. A 3D shape model of the left and right ventricles of the heart was constructed. This model contains both the average shape of these organs as well as their shape variability. The methodology presented in the last two chapters could also be applied to other anatomical structures. This has been illustrated in Chapter 6 with examples of geometrical models of the nucleus caudate and the radius.