Research On Segmentation For Magnetic Resonance Image Of Brain Based On Adaptive Graph Filter

Medical image segmentation is the basic step for three dimensional reconstruction and quantitative analysis of normal tissues and diseased tissues.MRI are widely used in brain detection due of the advantage of high spatial resolution and well soft tissue contrast.Since the MRI image adopts a parallel multi-coil acquisition method,the process of MRI imaging is affected by electronic noise and offset distortion,which result in noise,and biased field effects in the MRI image.In practical medical applications,due to the resolution restriction of MRI apparatus,there are volume effects in the MRI image.Volume effect is the boundary overlap between different organizations,the boundary voxels contains various tissues.Accurate segmentation can be challenging due to above problems.With the development of statistical learning,spectral graph theory and digital image processing technology,a novel method,which is discrete signal processing based on graph,provides new theories and methods for image segmentation problems.Clustering algorithm based on graph filter is the classic one in this theory.It is not a good choice in image segmentation to choose intensity of pixel or space feature to be the separate measures.However,graph signal method as a potential method can get the satisfied segmentation result by mapping the data to underlying graph,which leads the wise combination of intensity of pixel and space feature.There are also some flaws in Graph filter method.In order to correct these deficiencies,scholars have done a lot effort for the clustering algorithm,which also promotes the development theory of graph signal processing.Super-pixel method as a novel method can reduce the computational complexity of image processing effectively and suppress some noise.The super-pixel method focuses on the local area features.While enhancing the consistency of the image area,the method preserves the original boundary information of the image and the shape information of the area,which improves the accuracy of image segmentation.Therefore,it is widely used in the preprocessing stage of image segmentation in various fields.Based on this background,an effective algorithm for the brain tissue segmentation based on adaptive graph filter and super-voxels is proposed.The algorithm makes full use of the superiority of the graph signal and combines graph spatial information and pixel intensity information to obtain good results in experiments.Firstly,the algorithm divides the image by a super-voxel method to obtain a uniform size super-voxel.After that,the super-voxels is considered as graph vertex and mapped into a graph signal for smooth filtering to suppress the noise and bias field in the super-voxels.Finally,the segmentation results can be obtained through quadratic map.The method proposed in this paper can effectively overcome the influence of noise and bias field.Compared with the graph filter clustering method,it has higher segmentation accuracy and better robustness of noise.