Tissue Classification of Brain Structural Images by Using Bayesian Formulation with Python library and Comparing with Other Segmentation Methods

A very challenging part in the field of image processing is related to image segmentation, especially in the field of tissue classification. This issue will become much more problematic in the context of brain tissue segmentation in an MRI image, because brain tissue signal and contrast are much closer to each other. On the other hand, the type of programming language implementing these segmentation techniques is of great importance. In this study, the capabilities of Python libraries were used to classification brain MRI images. The technique used includes the implementation of the Bayesian formulation, where the observation model was defined as a Gaussian distribution, and a Markov random field (MRF) was used to model the prior probability of context-dependent patterns of different types of brain tissue. Finally, by using code written in Python language, T۱ weighted brain images were separated into three parts: gray matter, white matter and cerebrospinal fluid (CSF). Then two techniques, Fuzzy C-means clustering and mixture Gaussian model, were simultaneously implemented and the same segmentation were done. The results show well the good resistance of the MRF technique to noise and the implementation of a suitable segmentation in brain images. Since the noise and bias field is very serious, none of the methods can successfully perform the segmentation to achieve a perfect result. Two obvious weaknesses of FCM and mixture Gaussian is as follows. First, the classification of cortex and cerebellum is easily mistaken. Second, the segmentation seriously lacks spatial continuity. Compared with the other two methods, the MRF method can give us a continuous segmentation with the special basic assumption that the probability of a voxel depends on its neighborhood. Also, using the good flexibility of Python language, maximum and repeated conditional modes are used to find the optimal solution