A Multi-atlas Brain MR Image Segmentation Algorithm with Improved Sparse Representation

Abstract

Because macaque brain and human brain have a high similarity, human can deepen our understanding of human brain function by studying macaque brain tissue. In order to segment the subcortical nucleus of macaque brain more accurately, a multi-atlas brain MR image segmentation algorithm with improved sparse representation was proposed. First, the information of label atlas is introduced when the sparse representation image block is constructed; Then the mutual information is improved by changing the calculation method of information entropy, and used to measure the similarity between the target image and the atlas. These two measures make the weights of atlas more reasonable during fusion. In order to fuse the segmentation results of the fusion algorithm of the non-local patch base method and sparse representation method, a similarity index based on the combination of dice coefficient and cosine distance was proposed. The experimental results show that this algorithm improved the accuracy of segmentation of hippocampus, striatum, claustrum, and other nucleus, and has better robustness.

Keywords

• image segmentation
• mri
• macaque
• multi-atlas
• label fusion
• sparse representation
• mutual information