Unsupervised brain tissue segmentation in MRI images

During brain Magnetic Resonance Imaging (MRI) analysis, image segmentation provides information for the measurement and visualization of anatomical structures of the brain. Currently, segmentation performed by human experts is the gold standard method for such task, but it presents bias and variability dependence of the observer, due to issues as imaging device configurations, complex anatomical shape of tissues and captured noise. In this paper, we introduce a new unsupervised segmentation algorithm for brain tissue segmentation, which incorporates prior knowledge of the brain structure and 3D features of the image, to tackle some of these problems. To evaluate our algorithm, we built a synthetic brain MRI database of 20 subjects, which is also described here. Our algorithm obtained better performance than other three popular state-of-the-art methods.