Segmentation of Magnetic Resonance Brain Imaging Based on Graph Theory

Segmentation of brain images especially from magnetic resonance imaging (MRI) is an essential requirement in medical imaging since the tissues, edges, and boundaries between them are ambiguous and difficult to detect, due to the proximity of the brightness levels of the images.
In this paper, the graph-based method is proposed to solve the segmentation of MRI brain images wherein a weighted undirected graph is assigned to the image with each edge of the graph corresponding to an image pixel. The edge weight between two nodes demonstrated the similarity between two pixels of the image. Thereafter, a cost function, such as relative extremes and turning point, was assigned to the graph, which matched the derivation of the function. Minimization of this cost function, which was equivalent to the shortest path in a graph, led to image segmentation.
The advantageof the graph method over other methods is the simultaneous construction of spatial information at a high rate. Moreover, this method is implemented on the pixels in the space and can partition MRI brain images with low error in an effort to improve the previous methods. The comparisons demonstrated that the accuracy of the MRI brain image segmentation would be improved through the application of the present method.
The obtained results of the current study indicated the high accuracy of the proposed method (about 97.5%), compared to other similar methods. Therefore, this method can accurately distinguish various types of brain MRI tissues and have clinical applications.