Non-Rigid Image Registration for the Assessment of Myocardial Deformation from 3D Echocardiography (Niet-rigide beeldregistratie voor het bepalen van de hartspiervervorming met behulp van 3D ultrasone beelden)
Non-Rigid Image Registration for the Assessment of Myocardial Deformation from 3D Echocardiography
Cardiovascular diseases are currently the major cause of death in the wo rld. Cardiac imaging therefore plays an essential role in the diagnosis, management and follow-up of patients with any suspected or known cardio vascular disease. Ultrasound (US) imaging is a well established imaging modality in daily clinical practice for the evaluation of cardiac morpho logy and function by measuring cardiac wall motion and deformation (i.e. strain). Several recent technological advancements have made a real-tim e assessment of the heart in three dimensions now possible, and thus off er the possibility to improve and expand on the diagnostic capabilities of the traditional two-dimensional (2D) US images. Analyzing these datas ets however is a challenging endeavor as the spatial and temporal resolu tion is currently lower than in 2D and the large amount of data makes a manual evaluation cumbersome and subjective. The focus of the present thesis was therefore the development of non-rig id image registration techniques able to cope with these demanding condi tions. We illustrate that image registration is a viable technique for r egional cardiac function estimation by validating the technique in a var iety of cardiac ultrasound imaging scenarios. An in-vitro experimental s etup was built in which tissue-mimicking phantoms could be deformed and imaged. Some phantoms contained stiff inclusions to investigate to what extent dysfunctional areas could be identified. An in-vivo animal study was also designed to acquire volumetric data in 17 open-chest sheep subj ect to conditions comparable to those encountered in clinical situations such as ischemia. New regularisation methods were proposed to improve cardiac deformation estimates. A strategy was derived to adapt the topology of the control p oint grid of the non-rigid image registration method to the anatomy of t he heart. It was shown that such a model is more suited for cardiac def ormation. Efforts were made to automate the strain estimation workflow t o facilitate its introduction in clinical practice. Finally, excellent r esults were obtained at an international challenge where the performance of the proposed method was compared against the most competitive algori thms currently available.