Robust Patch-Based Multi-Atlas Labeling by Joint
Sparsity Regularization, Guorong Wu, Qian Wang +1 more
Abstract. Automated labeling of anatomical structures on MR brain images is widely investigated in many neuroscience and clinic studies to quantitatively measure either individual or group structural/functional difference. To address the issue of registration errors that affect multi-atlas based labeling, patch-based label fusion methods have been recently proposed by inspecting the similarity of each patch in the subject with all possible local patches from multiple atlases. Although the patch-based methods take the advantage of one-to-many correspondences to look for the correct matches for each subject point, it also takes the risk of including misleading local patches with different labels, which might undermine the final label fusion result. To alleviate this issue, we introduce the concept of sparsity in non-local label fusion by considering only a small number of patches from…