The Study On Tumor Segmentation For Brain MRI Image

Magnetic resonance imaging(MRI) has become an important supplementary means for the clinical diagnosis and treatment of brain tumor, owing to its non-invasive, multi-direction and multi-parameter imaging as well as the high resolution to soft tissue. Research on segmentation method for brain tumor in MRI image is not only the focus field of medical image segmentation, but also the realistic need of clinical application.According to the characteristics of brain MRI image such as gray inhomogeneity and fuzzy boundary, the strengths and weaknesses of current segmentation methods are systematically summarized. Making full use of the valuable information within images such as neighborhood grayscale, topological structure and spatial distribution, this research attempts to realize the accurate segmentation of tumor. Focusing on this theme, the main work of this paper is shown as follows:(1) This paper briefly introduces the common segmentation methods and evaluation indicators, as well as the basic concepts and operations of morphological image analysis. The important applications of mathematical morphology such as morphological filtering and watershed transform are expatiated in detail. Through the introduction of viscous morphological thoughts, multi-scale morphological modification is realized by simulating the flooding scenarios of viscous fluid.(2) The defects and improvement directions of fuzzy c-means(FCM) clustering algorithm are briefly analyzed. Considering that FCM is sensitive to gray inhomogeneity and noise interference, the spatial neighborhood information is introduced as a constraint. The types of pixels are distinguished by the local gray statistical information. Then, different pixels are modified by morphological closing operations with different-sized structuring elements. In this way, the local optimization and misclassification are avoided, meanwhile, achieving complete and accurate segmentation region. Finally, the segmentation results of proposed method are verified in terms of both visual quality and quantitative indices.(3) In consideration of the over-segmentation of watershed transform, in essence, it is due to the irregular details and noises within the image. In view of this, morphological hybrid opening and closing by reconstruction filtering is adopted to smooth and simplify the original image before constructing the gradient, while retaining the target contour information. Then, the gradient image is hierarchically modified in accordance with its three-dimensional topographic landform: processes the lower gradient layers with large-sized structuring elements, whereas the smaller-sized to the higher layers. Thus most local minimums caused by irregular details and noises are removed, while region contour positions corresponding to the target area largely remain unchanged. Finally, morphological watershed is employed to implement segmentation based on the multi-scale modified image. In addition, the influence of layering number on the final segmentation effect is analyzed. Experimental results show that, the suggested method could effectively restrain the over-segmentation of watershed transform, at the same time, retain its accurate contour localization.