Automatic alignment of intramodal tomographic data using s-distance approach

The alignment of volumetric datasets is an important problem in the processing of medical data. It is a prerequisite to numerous image based applications in diagnostic and therapeutic routines. In this paper, a new method is proposed for matching of 3D intramodality medical images. Our approach is based on some generalization of feature distance definition. Analogous to the standard surface matching, our algorithm uses also the chamfer distance like metric to define the quality of match function, however, the evaluation of the distance map is performed in a different way. The s-distance method is a step towards an automatic extraction of features, where each feature’s role in the registration process is weighted based on its relative statistical or spatial significance. As an alternative to the user-dependent non-automatic registration methods this approach offers a good assessment of similarity in the intramodality case. The elimination of less significant features in the registration process has resulted in a greatly improved efficiency over the voxel-based methods. Studying certain properties of the search space topography provides some insights into the performance of the proposed method as well as the standard registration algorithms in the rigid body registration problem.