MARS (Multi-Atlas Robust Segmentation) provides the automatic solutions for efficent segmentation/labeling anatomcial structures from medical images.

Specifically, this software has integrated several state-of-the-art multi-atlas based segmentation methods, such as majority voting, local weighted voting, and non-local patch based segmentation methods.

More importantly, we also included our recently-developed joint sparse patch based segmentation method in this software. Compared with convention methods, our method has the following advantages: (1) add sparsity constraint to suppress the influence of misleading patches; (2) reduce the joint risk of two patches jointly making the same segmentation errors, and (3) use iterative framework to correct the possible mis-segmentations.

This software package was developed in the IDEA group at UNC-Chapel Hill ( http://bric.unc.edu/ideagroup).
Wu et. al., "A generative probability model of joint label fusion for multi-atlas based brain segmentation", MIA, 2013.