MAP-based Brain Tissue Segmentation using Manifold Learning and Hierarchical Max-Flow regularization
Martin Rajchl, John S. H. Baxter, A. Jonathan McLeod, Jing Yuan, Wu Qiu, Terry M. Peters, James A. White, ALI RAZA KHAN · 2014
This document describes the methodology used for the team KSOM GHMF. The Advanced Segmentation Tools (ASeTs) include a maximum a-posteriori (MAP) segmentation method [1] to combine information from multi-atlas labeling with an intensity model. The intensity models for 7 structures (background, white matter, cortical gray matter, corticospinal fluid, ventricles, basal ganglia and white matter lesions) are learned from a training database (NTraining = 5) via manifold learning and subsequently combined and regularized using generalized hierarchical max-flow framework. The source code to the proposed framework will be made publically available on http://sourceforge. net/projects/ASeTs/