The need to compute an accurate spatial alignment between multiple representations of patient anatomy is a problem that is fundamental to many applications in computer assisted interventional medicine. One class of methods for computing such alignments is feature-based registration, which aligns geometric information of the shapes being registered, such as point-based landmarks or models of shape surfaces. A popular algorithm for feature-based registration is the Iterative Closest Point (ICP) algorithm, which treats one shape as a cloud of points that is registered to a second shape by iterating between point-correspondence and point-registration phases until convergence. In this presentation, a class of ``most likely point’’ variants on the ICP algorithm are presented that offer several advantages over prior ICP-based methods, such as high registration accuracy and the ability to confidently assess the quality of a registration outcome. The proposed methods are based on a probabilistic interpretation of the registration problem, wherein the shape alignment is optimized based on uncertainty models rather than minimizing the Euclidean distance between the shapes, as in ICP. This probabilistic framework is used to model anisotropic uncertainties in the measured shape features and to provide a natural context for incorporating additional features into the registration problem, such as orientations representing shape surface normals. The proposed algorithms are evaluated through simulation- and phantom-based studies, which demonstrate significant improvement in registration outcomes relative to ICP and other state-of-the-art methods.
Speaker Biography
Seth Billings is a PhD candidate in Computer Science at Johns Hopkins University, being advised by Dr. Russell Taylor and co-advised by Dr. Emad Boctor. Seth received his M.S.E degree in Computer Science from Johns Hopkins University and a B.S. degree in Electrical Engineering from Kettering University. Seth’s research at JHU has focused on systems and algorithms for computer assisted surgery. His primary research focus has been the development of feature-based registration algorithms that incorporate probabilistic frameworks for registering oriented feature data and for modeling anisotropic uncertainty in the features being registered. His research experience also includes development of an embedded multispectral light source to limit phototoxicity during retinal surgery, development of control software for the Laparoscopic Assistant Robot System (LARS), and development of a semi-autonomous surgeon-assistance mode for performing robot-assisted, laparoscopic, ultrasound-based elastography imaging with the da Vinci Surgical System. Seth also has prior experience as a full-time product engineer at the Research & Development division of Dematic, where he developed embedded systems for industrial automation following his graduation from Kettering University.