MRI segmentation of brain tissue based on spatial prior and neighboring pixels affinities

Segmentation of brain tissue in magnetic resonance imaging (MRI) can identify anatomical areas of interest. It has been widely used in medical imaging applications. In this paper, we propose a new brain MRI segmentation method, which takes advantage of spatial prior and neighboring pixels affinities. In addition, the underlying model can naturally describe the partial volume effects. Firstly, the algorithm labels some pixels based on pixel intensity and spatial prior. The label information is then propagated from the labeled pixels to unlabeled pixels with neighboring pixels affinities. Mathematically, the result is obtained by minimizing a quadratic objective function. In this way, we can extract different types of brain tissue from MR images and obtain the segmentation result. Experiments prove that our method can generate accurate results, which are comparable to that of the state-of-the-art methods.