Electronic Thesis and Dissertation Repository


Doctor of Philosophy


Biomedical Engineering


Terry M. Peters


Compared to conventional open heart procedures, minimally invasive off-pump beating heart mitral valve repair aims to deliver equivalent treatment for mitral regurgitation with reduced trauma and side effects. However, minimally invasive approaches are often limited by the lack of a direct view to surgical targets and/or tools, a challenge that is compounded by potential movement of the target during the cardiac cycle. For this reason, sophisticated image guidance systems are required in achieving procedural efficiency and therapeutic success. The development of such guidance systems is associated with many challenges. For example, the system should be able to provide high quality visualization of both cardiac anatomy and motion, as well as augmenting it with virtual models of tracked tools and targets. It should have the capability of integrating pre-operative images to the intra-operative scenario through registration techniques. The computation speed must be sufficiently fast to capture the rapid cardiac motion. Meanwhile, the system should be cost effective and easily integrated into standard clinical workflow.

This thesis develops image processing techniques to address these challenges, aiming to achieve a safe and efficient guidance system for off-pump beating heart mitral valve repair. These techniques can be divided into two categories, using 3D and 2D image data respectively. When 3D images are accessible, a rapid multi-modal registration approach is proposed to link the pre-operative CT images to the intra-operative ultrasound images. The ultrasound images are used to display the real time cardiac motion, enhanced by CT data serving as high quality 3D context with annotated features. I also developed a method to generate synthetic dynamic CT images, aiming to replace real dynamic CT data in such a guidance system to reduce the radiation dose applied to the patients. When only 2D images are available, an approach is developed to track the feature of interest, i.e. the mitral annulus, based on bi-plane ultrasound images and a magnetic tracking system. The concept of modern GPU-based parallel computing is employed in most of these approaches to accelerate the computation in order to capture the rapid cardiac motion with desired accuracy.

Validation experiments were performed on phantom, animal and human data. The overall accuracy of registration and feature tracking with respect to the mitral annulus was about 2-3mm with computation time of 60-400ms per frame, sufficient for one update per cardiac cycle. It was also demonstrated in the results that the synthetic CT images can provide very similar anatomical representations and registration accuracy compared to that of the real dynamic CT images. These results suggest that the approaches developed in the thesis have good potential for a safer and more effective guidance system for off-pump beating heart mitral valve repair.