A MAP based Approach Combining Intensity, Local Prior and Multi-atlas Prior for Brain Tissue Classification

Automated and accurate tissue classification in 3D brain Magnetic Resonance images is essential in volumetric morphometry or as a preprocessing step for diagnosing brain diseases. However, noise, intensity inhomogeneity and partial volume effects limit the classification accuracy of the existing methods. This work performs brain tissue classification using an approach combining three commonly used features: image intensity, local prior and multi-atlas prior. The image intensity is used by K-Means to obtain the initial classification. The local prior is modelled by a Markov Random Field to better deal with the images contaminated by severe artefacts. The multi-atlas prior is derived from applying exhaustive registrations and local label fusion strategy. In this work, we apply the multi-atlas segmentation method for brain segmentation. After that, a maximum a posteriori (MAP) approach combining intensity, local prior and multi-atlas prior is used to segment the brain tissues into three types: white matter, gray matter and cerebral spinal fluid.