ABSTRACT
In the present work a specific problem dependent solution has been introduced within the general mincut technique which is suggested by regularity of the spatial distribution of WM, GM, CSF. By exploiting the flexibility in the description of the energy function that allows to take into account a-priori information, work we consider a modified version of equation 2 by including a new term:
where A(L)= ∑v∈D Av(Ov) is an Atlas Term. We assume that each voxel v of the brain volume V has three a-priori cost values concerning the association of the specific voxel to theWhite Matter Av(OWM ), to the GrayMatter Av(OGM ) and to the Cerebral Spinal Fluid Av(OCSF ). These costs are derived using a brain tissue probabilistic atlas, made available from the LONI (Laboratory of Neuroimaging) of the UCLA University (Shattuck et al. 2008). The atlas consists of three volumes, which the value of each voxel indicates the probability that this belongs to the WM GM and CSF respectively. In the 3D lattice structure, besides the classic n-links and n-links, the a-links for each tissue are added. The weight of these connections is given by the a priori probability that the voxel belongs to a class, which is found in the probabilistic atlas. The min-cuts of this new graph are the 3D boundaries of the WM,GM anf CSF.
