Electronic Thesis and Dissertation Repository


Doctor of Philosophy


Biomedical Engineering


Dr. Terry Peters


The introduction of transcatheter aortic valve implantation (TAVI), an innovative stent-based technique for delivery of a bioprosthetic valve, has resulted in a paradigm shift in treatment options for elderly patients with aortic stenosis. While there have been major advancements in valve design and access routes, TAVI still relies largely on single-plane fluoroscopy for intraoperative navigation and guidance, which provides only gross imaging of anatomical structures. Inadequate imaging leading to suboptimal valve positioning contributes to many of the early complications experienced by TAVI patients, including valve embolism, coronary ostia obstruction, paravalvular leak, heart block, and secondary nephrotoxicity from contrast use.

A potential method of providing improved image-guidance for TAVI is to combine the information derived from intra-operative fluoroscopy and TEE with pre-operative CT data. This would allow the 3D anatomy of the aortic root to be visualized along with real-time information about valve and prosthesis motion. The combined information can be visualized as a `merged' image where the different imaging modalities are overlaid upon each other, or as an `augmented' image, where the location of key target features identified on one image are displayed on a different imaging modality.

This research develops image registration techniques to bring fluoroscopy, TEE, and CT models into a common coordinate frame with an image processing workflow that is compatible with the TAVI procedure. The techniques are designed to be fast enough to allow for real-time image fusion and visualization during the procedure, with an intra-procedural set-up requiring only a few minutes. TEE to fluoroscopy registration was achieved using a single-perspective TEE probe pose estimation technique. The alignment of CT and TEE images was achieved using custom-designed algorithms to extract aortic root contours from XPlane TEE images, and matching the shape of these contours to a CT-derived surface model. Registration accuracy was assessed on porcine and human images by identifying targets (such as guidewires or coronary ostia) on the different imaging modalities and measuring the correspondence of these targets after registration.

The merged images demonstrated good visual alignment of aortic root structures, and quantitative assessment measured an accuracy of less than 1.5mm error for TEE-fluoroscopy registration and less than 6mm error for CT-TEE registration. These results suggest that the image processing techniques presented have potential for development into a clinical tool to guide TAVI. Such a tool could potentially reduce TAVI complications, reducing morbidity and mortality and allowing for a safer procedure.