Joint image segmentation and registration based on a dynamic level set approach using truncated hierarchical B-splines

We present a novel approach for joint image segmentation and nonrigid registration using bidirectional composition based level set formulation. This efficient framework incorporates automatic structural analysis from image segmentation into the registration framework. This method has shown an improved performance as compared to carrying out segmentation and registration separately. Unlike previous approaches, the implicit level set function defining the segmentation contour and the spatial transformation function that maps the deformation for image registration are both defined using B-splines. This joint level set framework uses a variational form of an atlas-based segmentation together with large deformation based nonrigid registration. In addition, a bidirectional composition framework is introduced to incorporate a more symmetric update. The minimization of the variational form is accomplished by dynamic evaluations on a set of successively refined adaptive grids at multiple image resolutions. The improvement in the description of the segmentation result using higher order splines leads to a better accuracy of both the image segmentation and registration process. The performance of the proposed method is demonstrated on synthetic and medical images to show the improvement as compared to other registration methods.