Inverse consistent registration with object boundary constraints

This paper describes our preliminary work on developing a new inverse consistent image registration algorithm that registers objects of interest in the images rather than the entire image. It is assumed that the boundary of an object or region-of-interest (ROI) in the template and target images are given as input to the image registration algorithm. The object/ROI boundaries and the intensity inside the boundaries are jointly registered while minimizing the inverse consistency error. The inverse consistency error is minimized when the forward and reverse transformations between two images are inverses of each other. This new registration algorithm is formulated in a Eulerian coordinate system giving a common coordinate system representation for both the boundary and regional parameterizations of the transformation. This new algorithm provides computational advantages compared to matching entire images and provides an efficient means to incorporate boundary conditions into intensity driven image registration.