A New Brain Segmentation Framework

We present a new brain segmentation framework which we apply to T1-weighted magnetic resonance image segmentation. The innovation of the algorithm in comparison to the state-of-the-art of nonsupervised brain segmentation is twofold. First, the algorithm is entirely non-parametric and non-supervised. We can therefore enhance the classically used gray level information of the images by other features which do not fulfill the parametric Gaussian assumption. This is illustrated by a segmentation algorithm that considers both, voxel intensities and voxel gradients for the segmentation task. The resulting algorithm is called a non-supervised, non-parametric hidden Markov random field segmentation. Furthermore we have also to construct an anatomically relevant segmentation model in the resulting two-dimensional feature space. This is the second main contribution of this paper. We construct a morphologically inspired classification model, which is also able to segment the deep structures of the brain into a separate class, resulting in a six class segmentation model. We prove the validity of the introduced mathematical and morphological aspects on simulated T1-weighted magnetic resonance images of the brain.