Grey matter segmentation of 7T MR images

Abstract

The aim of this research is to produce an accurate segmentation of the brain grey matter tissue of a 3D MR (Magnetic Resonance) image from a high field (7T) MR scanner. 7T scanners produce images with a high SNR (Signal to Noise Ratio), but also with high inhomogeneity that makes brain segmentation a very challenging problem. The level set method is a popular method for image segmentation. However, the inhomogeneity of voxel intensities within the same tissue types lead to inappropriate placement of seed points and parameter initialisation, and therefore poor segmentation results. To overcome the inhomogeneity problem we combine a statistical modelling approach, taking inspiration from the method implemented in the SPM software package, and the level set segmentation method. We first correct inhomogeneity by multiplying the image voxels with a bias field (modulation image). The level set method is then used to segment the bias corrected image. The resulting probability maps from the statistical model are also incorporated into the level set segmentation to improve initial seed selection and level set surface evolution.