Multi-resolution elastic registration of CT and MR brain images

A key issue for multi-modality medical image processing is to match images from different resources. This paper describes a multi-resolution method to perform elastic matching of two medical images. First, a topologically adaptable snake is used to extract the geometrical description of salient anatomical feature in both images. Then a global transformation is computed to match two images globally. After that, a matching points selecting method is proposed, which can automatically choose matching points based on resolution distance. This distance is changed from large to small step by step and is used to determine the grid size in multi-resolution analysis. The smaller the distance, the finer the image level and the boundary detail determined by the selected feature points. In each resolution level, based on the external force vectors produced by the newly selected matching points, an elastic warping transformation is then determined, which brings two images into registration. Hence one can register the two images step by step, increasing the local similarity until reach the required accuracy.