Image-based Tracking Methods - Applications to Improved Motion Compensation in Cardiac MR and Image-Guided Surgery

In this dissertation, we present our efforts in developing algorithms in three related areas: 1) unified and optimal tracking, 2) improved motion compensation in cardiovascular MR, and 3) tool tracking in image-guided surgery. With the rapid advancement in the development of tracking algorithms over the years, the choice of tracking methods for a particular application has significantly increased, but it is generally not clear what is the best algorithm to use. Recently popularized sampling kernels have emerged as a flexible design space for developing and merging various tracking algorithms. We demonstrate a connection between kernel-based algorithms and more traditional template tracking methods. This allows for a more efficient optimization (convergence in fewer steps) and naturally extends to objective functions optimizing more complex motion models using multiple spatially distributed kernels. The efficacy of the multi-kernel methods is demonstrated on a variety of tracking examples. In conventional kernel-based tracking approaches, the kernel parameters are selected in an ad-hoc fashion usually leading to sub-optimal tracking results. Thus, we present results pointing toward the design of optimal and approximately optimal target-specific kernels for tracking. In image-guided robot-assisted eye surgery, the dynamic reconstruction of the surgical field that involves the surgical tools and retinal surface and