We propose a novel multi-atlas segmentation method that employs a group-wise image registration method for the brain segmentation on rodent magnetic resonance (MR) images. be naturally extended to a novel multi-atlas segmentation method and improves the registration method to explicitly use the provided template labels as an additional constraint. In the experiment we show that our segmentation algorithm provides more accuracy with multi-atlas label fusion and stability against Magnolol pair-wise image registration. The comparison with previous group-wise registration method is provided as well. anatomical information is crucial to simplify the segmentation process. In atlas-based segmentation an intensity image Magnolol with a segmentation label is provided to be matched to the target image we wish to segment and the segmentation result is then obtained by warping the label image to the target image via the transformation field estimated by the deformable image registration. Sophisticated image registration methods 2 such as Magnolol Large Diffeomorphic Deformation Metric Mapping (LDDMM) have greatly helped to solve the segmentation problem but depending on anatomical variation among the population and characteristics of the registration method used an atlas-based segmentation method in particular using a single atlas cannot always produce accurate results. For Magnolol example the segmentation result may be biased to the choice of atlas may not be robust enough to perform a group analysis and may not be stable in the presence of irrelevant structures such as non-brain voxels. Group-wise approaches that simultaneously consider multiples of atlases have received more and more attention in recent years due to their importance in population analyses. Group-wise approaches in atlas-based segmentation can be classified into two categories: 1) multi-atlas label fusion; and 2) group-wise image registration. Using a pair-wise image registration Rabbit Polyclonal to Thyroid Hormone Receptor beta. between an atlas and the target image multi-atlas label fusion method makes use of more than one atlas to mediate potential bias associated with using a single atlas and applies label fusion method to create the final segmentation. This method requires additional computational costs to independently perform image registration per each atlas but several empirical studies have recently shown that the method is more accurate than single atlas-based segmentation. Most existing label fusion methods are based on a weighted voting scheme where each atlas contributes to the final segmentation according to a non-negative weight and atlases more similar to the target image yield larger weights. Among weighted voting methods those that derive weights from local similarity between the atlas and target have been most successful in practice allowing the weights to vary spatially. In contrast to multi-atlas label fusion the use of group-wise image registration helps to improve registration performance by simultaneously registering a group of images rather than a pair of images. For instance the use of a population atlas has shown to be more robust than using only external atlases since the population atlas can capture the variation of a population more closely than the external one. The creation of an unbiased population atlas based on diffeomorphic deformable registration is originally proposed by Joshi et al.3 Meanwhile Hamm et al.4 presented a tree-based registration method for group-wise registration such that the intrinsic anatomical manifold is learned by a manifold learning technique and represented as a tree rooting from the geodesic mean image. The path between two images on the tree provides a series of small deformations so that it can guide large deformation registration which is hardly successful for pair-wise registration. These methods improve the accuracy of registration by considering a group of images but they are fundamentally based on iterative pair-wise registrations or using the similarity metric derived from pair-wise registration. Rather than using a pair-wise registration the proposed segmentation method applies a group-wise registration that evaluates a group-wise similarity metric and gradually transforms all subjects toward an implicitly defined common reference frame. The use of a group-wise similarity.