Simultaneous Brain Structures Segmentation Combining Shape and Pose Forces

This paper presents a new supervised learning based method for brain structure segmentation. We learn moment-based signatures of structures of interest and formulate the segmentation as a maximum a-posteriori estimation problem employing nonparametric multivariate kernel densities. For this problem, we propose a gradient flow solution. We have compared our method with state-of-the-art methods such as FSL-FIRST and Free-Surfer using volumetric 3T from IBSR. In addition, we have evaluated our algorithm on 7T MR data. We report comparative results of accuracy and significantly improved time-efficiency.