Automatic Caudate Segmentation by Hybrid Generative/Discriminative Models
Zhuowen Tu, M. Berk Mirza, Ivo D. Dinov, Arthur W. Toga · 2007
Abstract. Segmenting sub-cortical structures from 3D brain images is of significant practical importance. This paper presents an experimental study for caudate segmentation in MRI images using a fully automatic algorithm based on [9], which is an extension of the hybrid model ap-proach described in [1]. The method in [9] tackles multiple sub-cortical and cortical structures. In this study (the grand challenge competition), the focus is on segmenting the left and right caudate from MRI images. Given a set of training data, a classifier (PBT [9]) is used to learn a dis-criminative model for each voxel being on the left caudate, right caudate, or the background. In testing, a variational approach is used to perform segmentation by minimizing an energy of a hybrid model. This hybrid model is composed of a discriminative model term by the trained PBT classifier and a generative shape term. It takes, typically, 5 minutes to perform the segmentation on a modern PC. 1