3D segmentation of the cerebrospinal fluid from CT brain scans using local histogram similarity map

Abstract

This paper considers the problem of segmentation of regions filled with the cerebrospinal fluid (CSF) from CT brain images. In particular a 3D segmentation method is proposed. The method performs segmentation by comparing intensity distribution within the user defined seed region with local intensity distributions around all image voxels. In order to speed up the computations, GPU computing is used. Results of applying the introduced approach to segmentation of CSF accumulated in different brain regions are presented and discussed. Issues regarding accuracy and execution time of the method are also taken into account.