Abstarct: MRI brain image segmentation plays an increasingly important role in computer-aided detection and diagnosis (CAD) of abnormalities. This thesis uses MRF as an unsupervised method for MRI segmentation. MRF is a statistical method that poses image segmentation as a labeling problem which seeks an optimal label field in a large solution space. In order to speed up the algorithm, we have proposed two novel optimization methods. The first combines Simulated Annealing (SA) and an Improved Genetic Algorithm (IGA). SA is a powerful searching method which imposes heavy computation burden on the algorithm. In comparison, genetic algorithm (GA) has a good capability of searching which converges quickly to a near-global optimum. By combining SA and IGA for optimization, this project puts forward a new model for segmentation which keeps the benefits of both while simultaneously addressing their individual drawbacks. The next, uses social algorithms contained ant colony optimization (ACO) and gossiping algorithm. ACO is a multi-agent guided method for optimization which is inspired from foraging behavior of ants in order to find optimum path between nest and food resource. Tackling Gossiping algorithm, the combinational method assists ants in smart decision. Therefore, the proposed algorithm outperforms the classical MRF model in speed and quality of the solution. In this thesis, we have used three MRI datasets to compare two proposed methods with the classical MRF. First, the algorithms are used to segment normal MRI brain tissues contain White Matter (WM), Gray Matter (GM), and Cerebrospinal Fluid (CSF). For classical MRF results indicate the average value of Dice coefficient percentage 72.1%. This measurement for MRF-SA-IGA is equal to 73.1% and For MRF-ACO-Gossiping is 72.7%. Also the average of computational time reduction for MRF-SA-IGA in compare with classical MRF is 89.49%. In addition, this parameter for MRF-ACO-Gossiping in compare with classical MRF is equal to 59.02%. Next, the algorithms are used to segment tumors in real MR images. The average value of Dice coefficient percentage for classical MRF, MRF-SA-IGA, and MRF-ACO-Gossiping are equal to 73.10%, 76.00% and 70.80% respectively. Also computational time reduction average of MRF-SA-IGA in compare with classical MRF is 91.33%. Additionally this parameter for MRF-ACO-Gossiping in compare with classical MRF is equal to 71.21%. Last, the algorithms are used to segment multi-spectral MRIs into BG, WM, GM and CSF. For classical MRF results indicate the average value of Dice coefficient percentage 82.1%. This value for MRF-SA-IGA and MRF-ACO-Gossiping are equal to 83.2% and 82.7% respectively. Also computational time reduction average of MRF-SA-IGA compare with classical MRF is 77.08%. This parameter for MRF-ACO-Gossiping compare with classical MRF is equal to 67.08%.