Image Registration Between X-ray Fluoroscopy and Transesophageal Echocardiography: Novel Algorithms and Applications for Transcatheter Aortic Valve Replacement
Since it was first performed in humans in 2002, transcatheter aortic valve replacement (TAVR) has emerged as a successful minimally invasive treatment for aortic valve disease. In TAVR, an artificial prosthetic valve is deployed by a catheter, typically under guidance of x-ray fluoroscopy (XRF). Because proper valve positioning is important for achieving optimal clinical outcomes, advanced image guidance systems and workflows that use all available imaging modalities may be able to further improve the success of this procedure. Recently, image registration between XRF and transesophageal echocardiography (TEE) has been validated and clinically implemented (Philips EchoNavigator). This technology uses image processing to merge the XRF and TEE coordinate systems, allowing the information from both modalities to be fused into a single visualization framework. It is hypothesized that image guidance during TAVR can be improved using XRF/TEE registration by allowing anatomical information from TEE to be combined with device visualization from XRF. In this thesis, technical contributions aimed at enhancing image guidance and clinical workflows for TAVR using XRF/TEE registration are presented. In the introductory chapter, the history of interventional cardiology is briefly discussed, followed by a description of the TAVR procedure. Novel clinical workflows aimed at improving procedural efficiency and prosthetic valve deployment accuracy are proposed, and specific technical problems involving the proposed workflows are identified and addressed in chapters 2-5 In chapter 2, a novel implementation of the Hough forest algorithm for object detection is presented and applied to the problem of automatic TEE probe and prosthetic valve detection in XRF images. The purpose of this aim is to minimize the need for user interaction in the image registration process, enabling enhanced clinical workflows and image guidance. In clinical datasets from 48 patients, the TEE probe was successfully detected in 95.8% of images (n=1077) and the prosthetic valve was detected in 90.1% of images (n=388). These results indicate that the presented method is feasible and has potential for clinical use. Along with a summary of the technical background and prior work concerning XRF/TEE registration, chapter 3 presents two novel algorithms designed for improved registration accuracy and speed. In chapter 4, these algorithms were validated in simulated, phantom, and clinical datasets. It is shown that the first proposed algorithm was an order of magnitude faster and had a higher success rate than state-of-the-art methods, but was slightly less accurate. The second proposed algorithm was faster and more accurate than state-of-the-art methods, but had a lower success rate. When both algorithms were combined in a hybrid approach, state-of-the-art methods were greatly outperformed in all categories, leading to the first method for XRF/TEE registration that is purely image based (requires no extra hardware), accurate, and fast enough to operate at fluoroscopic frame rates. In chapter 5, a clinical application of XRF/TEE registration is introduced. A method for contrast-free, intraprocedural optimization of TAVR projection angles using XRF/TEE registration was developed and tested in 10 patients. It was shown that the proposed method agreed with the standard aortographic method to within 3.46 +- 3.28 deg., while a previously introduced method using preoperative CT agreed to within 7.01 +- 2.78 deg. Furthermore, the proposed method can be performed intraoperatively with minimal disruption of clinical workflow and without the use of nephrotoxic x-ray contrast dose. In the final chapter of this thesis, the potential impact of the presented algorithms are discussed in the context of future image guidance systems for TAVR. Limitations of the current methods and future work needed for clinical translation of the proposed technology are discussed.