Localization and Mapping for Guiding Minimally Invasive Cardiac Procedures.

To successfully perform minimally invasive surgical procedures, physicians must be able to visualize their instruments inside the anatomy. Unfortunately, guidance facilities for several such procedures are limited; this is especially true in the cardiac setting. The two imaging modalities which provide the most-detailed views of the anatomy---computed tomography (CT) and magnetic resonance (MR)---are presently not well-suited for guiding cardiac interventions. Ultrasound, by contrast, provides real-time imaging without radiation exposure, but its field-of-view and ability to discriminate between tissues are limited. This dissertation presents a set of novel registration algorithms that fuse intracardiac ultrasound, pre-operative CT/MR, and electromagnetic position sensing data to create a "virtual surgical environment," allowing physicians to visualize their instruments inside a detailed surface mesh model of the anatomy. These algorithms are based on principles taken from robotic localization and mapping---a well-developed set of techniques used to locate mobile robots within and construct maps of their environment. Here, the mobile robot is an ultrasound imaging probe tracked by a position sensing system, and its environment is the heart. By localizing the ultrasound probe inside the heart, other instruments tracked by the position sensing system are localized as well. The ultrasound probe is then used to maintain localization for the duration of the procedure. We evaluate the performance of these algorithms in three clinically-relevant scenarios. First, we assume that an anatomic surface mesh whose shape matches that of the intraoperative anatomy can be obtained by segmenting pre-operative CT/MR; the surface mesh is then used as a "reference map" while the ultrasound probe "navigates" inside the heart. Localization within the reference map is equated to localization within the anatomy because of their assumed identical structure. Next, we relax our assumption that the pre-operative surface mesh and intra-operative anatomy are identically-shaped. This scenario is more realistic because pre-operative image data is typically "out-of-date" at the start of the procedure. We localize the ultrasound probe within the outdated reference map, and use acquired ultrasound data to either construct a new map, or update the outdated map. Finally, we assume that pre-operative CT/MR was not acquired. A reference map must be created from scratch during the procedure while simultaneously localizing the ultrasound probe within the map. Our algorithms are fast, automatic, and accurate, ensuring that tissue contact in the anatomy is properly indicated inside the virtual model. Furthermore, ultrasound probe localization is updated continuously to compensate for dynamic phenomena that occur during the procedure, including patient motion. In achieving the above, we have created a catheter guidance system that enables cardiologists to operate in an accurate virtual surgical environment which is created rapidly, robust to real world perturbations, and equally easy-to-visualize as three-dimensional CT and MR.