Multi-Step Procedures for the Localization of 2D and 3D Point Landmarks and Automatic ROI Size Selection

Abstract

In this contribution, we are concerned with the detection and refined localization of 3D point landmarks. We propose multi-step differential procedures for subvoxel localization of 3D point landmarks. Moreover, we address the problem of choosing an optimal size for a region-of-interest (ROI) around point landmarks. That is, to reliably localize the landmark position, on the one hand, as much as possible image information about the landmark should be incorporated. On the other hand, the ROI should be restricted such that other structures do not interfere. The multi-step procedures are generalizations of an existing two-step procedure for subpixel localization of 2D point landmarks. This two-step procedure combines landmark detection by applying a differential operator with refined localization through a differential edge intersection approach. In this paper, we investigate the localization performance of this two-step procedure for an analytical model of a Gaussian blurred L-corner. The results motivate the use of an analogous procedure for 3D point landmark localization. We generalize the edge intersection approach to 3D and combine it with 3D detection operators to obtain multi-step procedures for subvoxel localization of 3D point landmarks. The multi-step procedures are experimentally tested for 3D synthetic images and 3D MR images of the human head. We also propose an approach to automatically select an optimal ROI size. This approach exploits the uncertainty of the position estimate resulting from the edge intersection approach. We present first promising results for a real 2D image with different types of corners as well as for anatomical brain landmarks in 2D slices of a 3D MR image.