The Brain MRI Segmentation Technology Research Based On Improved Graph Cut Algorithm

Image segmentation is the under-Layer technology in iconography, and it lays the foundation for computer vision research and depth image analysis. In recent years, with the continuous expansion and development of the medical field, a variety of medical equipment has also been the emergence, which provides a large amount of image data for clinical medicine. To use image segmentation technology for assist physicians quantitative analysis of medical images, thereby improving the diagnosis and to develop diagnostic protocols, many scholars will combine computer vision with clinical medicine. Wherein the human brain as the main core organ, a lot of mental illness(mental sluggishness, senile dementia, attention deficit hyperactivity disorder and paranoia, etc.) are closely related with some organizational structure of the brain, therefore, to quickly and accurately extract these organizational structures and developing a way tools and techniques becomes urgent.In the past few decades, graph cut optimization technology because of whose scalable and fast, and it will be welcomed and applied to image segmentation by the majority of researchers. Because of the complexity of brain MRI organizational structure, and gray between the organizations has some particularity characteristics of the uneven distribution and low contrast, traditional graph cut algorithm for segmentation may produce situation of shrinking bias and border local optimum, so in order to achieve better segmentation of brain tissue structure, and the paper studies the graph cuts theory, what based on traditional graph cut algorithm made some improvements.In order to overcome this phenomenon that target border is prone to misclassification, when operator selects fewer pixels seed point in the original graph cut algorithm, In this paper, we propose the KMGC algorithm based on combine the k-means with GC(Graph Cut) algorithm to interactive segmentation with Brain Magnetic Resonance Image(MRI). the MRI intensity inhomogeneity were processed by k-means clustering algorithm, on this basis, graph cut algorithm will further refine the MRI, so as to obtain effective segmentation of white matter and gray matter. In this paper, not only for the white matter and gray matter segmentation, but also in the deep gray matter nucleus(caudate nucleus) tissue segmentation, but since the caudate nucleus in the gray matter has characteristics of low contrast and fuzzy boundaries, etc., and we want to take them extract from gray matter, using only gray information is not enough, this paper proposes a method based on adaptive fuzzy connectedness combined with graph cut(AFCGC) of the caudate nucleus is divided, and algorithms of chapters III and IV of this paper respectively compared with other algorithms in qualitative and quantitative analysis to verify the proposed algorithm in the segmentation result is better than other algorithms.In this paper, the main research work as follows:1) the MR image segmentation and evaluation methods of segmentation algorithms are summarized.2) on the graph cut techniques were studied and noted that shortcoming of traditional graph cut algorithm and improved methods.3) In this paper, we proposed and verified KMGC algorithm for the effective segmentation of brain gray natter and white matter.4) we proposed and verified KMGC algorithm for the effective segmentation of the caudate nucleus.