Research On Intracranial Hemorrhage Region Segmentation Based On Supervoxel

Intracranial hemorrhage(ICH)is one of the most serious cerebrovascular diseases and also is an important cause of acute neurological disorders(hemiplegia).For clinical treatment,the segmentation of intracranial hemorrhage is of great significance.Therefore,how to use CT images to diagnose intracranial hemorrhage has become one of the most popular research topics in the field of brain medicine.In traditional medical image analysis,medical practitioners rely mainly on their own experience and manual mapping to obtain an illness judgment.The introduction of image segmentation technology greatly reduces the burden of medical practitioners.Quantitative results obtained by segmentation provide medical practitioners with an accurate basis for diagnosis.In recent years,a variety of image segmentation algorithms have been proposed.Because of its good segmentation performance,algorithms based on superpixel,graph theory and semi-supervised learning have become the focus of attention of many researchers.The superpixel algorithm divides the image into small regions by using the similarity of pixel features,that reduces the redundant information and the complexity of subsequent image processing.Image segmentation algorithm based on graph theory transforms the image segmentation problem into the network graph segmentation problem by associating the image characteristics with the graph-theoretic properties.By combining the global segmentation and the local information processing,the image discretization error is reduced and good segmentation results can be obtained.In the case of scarcity of labeled data,image segmentation based on semi-supervised learning can use a large amount of unlabeled data to enhance segmentation results.This paper focuses on the study of intracranial hemorrhage regional segmentation and the characteristics of superpixel,especially on the application of superpixel-based graphcut algorithm and superpixel-based Tri-training algorithm in intracranial hemorrhage regional segmentation.The main contributions of this paper are listed as follows:1.The algorithms used in image segmentation and their respective application range are studied.The characteristics of medical image segmentation and imaging characteristics of intracranial hemorrhage(ICH)are briefly introduced.2.The basic principle,advantages and disadvantages of superpixel are introduced in detail.Based on the existing superpixel algorithm and the application of intracranial hemorrhage region segmentation,a new supervoxel algorithm is proposed.3.Aiming at solving the problem of human participation and lack of model estimation in graphcut algorithm,a supervised graphcut algorithm based on Gaussian Mixture Model(GMM)is proposed.According to the characteristics of medical images,the GMM algorithm is used to build the foreground and background models based on the prior knowledge of labeled data,so that the image segmentation algorithm based on graphcut can realize automatic segmentation.4.The basic principle,advantages and disadvantages of Tri-training algorithm are introduced in detail.An algorithm of intracranial hemorrhage regional segmentation based on supervoxel and Tri-training algorithm is proposed.The algorithm realizes the automatic segmentation of medical images by using a small amount of labeled data and a large amount of unlabeled data.