An Eigenvector-based Corresponding Points Auto-Detection Algorithm for Non-rigid Registration of CT Brain Images

CT brain images have advantage on the detection of brain tumor and cerebral hemorrhage. To provide integrated scheme of computer-aided diagnosis of these diseases with CT images, the non-rigid registration of CT images should be researched. In this paper, we present an eigenvector-based corresponding point automatic detection for non-rigid registration of CT brain images. Geometrical moment is first constructed for each voxel, and subsequently standard deviation of geometrical moment is computed. Eigenvectors, based on standard deviation and edge strength, are used to automatically and exactly determine the corresponding landmark points on fixed image for the pre-labeled landmark points on moving image and vice versa. The experiments show that the eigenvector-based point match method is accurate. It will be incorporated in the semi-auto non-rigid registration scheme for the further research.