Research On Brain Tumor Segmentation Of MR Image

Magnetic Resonance Imaging(MRI)is an important medical diagnostic tool,because of its advantages of high tissue contrast and resolution,of higher sensitivity and non-radiation to morphological and pathological changes in tissues,which now have been widely applied to detect pathological lesions and diseases in various tissues and organs,especially in tumor detection.MR brain tumor image provides a very good help for clinical diagnosis of medical personnel and formulate treatment.However,how to segment MR images of brain tumors is not only an emphasis and difficulty of the clinical diagnosis and treatment,but also one of MRI brain tumor image processing and analysis.Based on the existing research of MRI brain tumor segmentation algorithms,this thesis proposed based on improved Watershed algorithm,Mean Shift algorithm and Max-flow/Min-cut algorithm to achieve the segmentation of MRI brain tumor images.First of all,this thesis analyzes the characteristics of MRI and the difficulty in brain tumor segmentation,also briefly describes the theory of many algorithms in medical image segmentation,as well as their advantages and disadvantages in MRI tumor segmentation.These laid a good foundation for follow-up study.Secondly,the segmentation algorithm of MR images based on watershed algorithm is studied.The advantage of watershed algorithm is that the precise contour of single pixel width and closed connected single pixel can be obtained,while its disadvantage is prone to over segmentation.In order to well restrain the over segmentation,the morphology opening and closing by reconstruction filter are firstly adopted,and then the foreground object and the background object were labeled using marker controlled method,and finally watershed segmentation is put to use.The improved watershed algorithm effectively solves the over segmentation problem.Again,the thesis focuses the segmentation algorithm based on Mean Shift of brain tumors in MR images.The Mean Shift algorithm is a segmentation method based on region.The algorithm has stronger robustness and adaptability,but it in image segmentation process also exist some problems,such as over segmentation and under segmentation,operating speed and segmentation efficiency,etc.To solve these problems,this thesis introduces a new iterative step,which combined with the existing acceleration strategy,to put forward a kind of improved accelerated algorithm,in which,region merging is performed by maximum entropy,with a smaller region segmentation,finally make the brain tumor segmentation more accurate.Finally,this thesis studies the theoretical foundation and implementation framework of image segmentation based on Graph Cuts theory at first,and then using the maximum flow minimum cut theorem for image segmentation.Because of the traditional graph cut needs construct a lot of graph nodes and local consistency is not strong,the efficiency of the algorithm is very low.In order to solve the problem,this thesis construct a new foreground/background probability model to improve operational efficiency and achieve accurate image segmentation.