Brain scanning using Magnetic Resonance Imaging (MRI) can be used to detect the brain tumor. MRI could detect the soft tissue abnormalities better than the other radiological devices. However, the noise in the image of the MRI sometimes appears randomly, so that it is difficult to detect the tumor more precisely. The image segmentation, therefore, is needed to be able to diagnose the location of the brain tumor by separating the tumor as the Region of Interest (ROI) from other regions. Gaussian Mixture Model (GMM) is commonly used for image segmentation. This method, however, frequently provides a poor result since it is less able to explain the skew pattern of MRI data. Moreover, the GMM is not considering the spatial dependencies between pixel, therefore it is less capable of handling noise. This study tries to employ the Fernandez Steel Skew Normal (FSSN) distribution as the replacement of the Gaussian in the GMM. The FSSN distribution could accommodate symmetrical and even asymmetrical pattern of the MRI data adaptively. In order to increase the noise robustness, the spatial dependencies with Markov Random Field (MRF) are used as a prior in Bayesian Markov chain Monte Carlo. The proposed model is called as the Spatially Constrained FSSN mixture model (Sc-FSSNMM). The results show that by applying the Sc- FSSNMM, the segmentation result is better than the GMM. In additions, the Sc-FSSNMM is more robust to noise while also more parsimony.