||Image registration is the process to spatially align corresponding structures in two images. In this presentation I will mainly talk about image registration as a segmentation method, atlas-based segmentation, but I will also briefly discuss other usages of image registration in radiotherapy, such as dose accumulation and automatic positioning. When used as a segmentation method, results can be improved by combining results from several image registrations, a process known as label fusion. I will present such a label fusion method, probabilistic weighting. We use a linear combination of signed distance maps as segmentation representations, and probabilistic weighting calculates weights for the linear combination incorporating image similarity information from the deformed atlas images by learning correlations between image similarity and segmentation quality.