Atlas based segmentation by diffeomorphic deformation

In this paper, we present a new framework for multiple objects segmentation based on deformable template. In medical applications, a reference atlas is often available which shows the topology of different structures and their relative positions. We propose to evolve the atlas image under diffeomorphic deformations to match the structures in the image to be segmented. The topologies and relative positions between different objects are maintained during the deformation. Finding the optimal mapping is formulated as a variational problem over a vector field. A gradient descent based multi-resolution algorithm is detailed, to reduce both ambiguities and computation load. This is illustrated by numerical experiments on various synthetic and real images. We expect this method be a useful tool for analysis medical images and other images which have similar properties.