Research On Segmentation Methods Of Cervical Cytology Image

The accurate segmentation of cervical cell images is one of the key steps of the cervical cancer computer-aided diagnosis system.In recent years,the cervical cell image segmentation technology has made significant progress,and the performance of the cervical cancer diagnosis system has been greatly improved.However,due to the characteristics of cervical cell images and the complexity of cell morphology,the accuracy of the existing cervical cell image segmentation technology still needs to be improved.Based on the application background of assisted diagnosis of cervical cancer,this paper deeply researches the identification and segmentation of cervical cell images.The main research results of the thesis are as follows.In the cell cluster segmentation of cervical cell images,a cell cluster segmentation method based on multi-scale graph cut algorithm is proposed.The multi-scale graph cut algorithm and the confidence region method are used to obtain the global seed node,and the probability distribution map of pixel is generated for the global graph cut algorithm in the confidence region.According to the global seed node,the probability distribution map and the global graph cut algorithm,the cell cluster region segmentation in the cervical cell image is completed.Compared with the threshold algorithm and the watershed algorithm,the cell cluster segmentation algorithm based on multi-scale graph cut in this paper can accurately segment the cell cluster area in the complex background.In terms of cell nucleus segmentation,a cervical cell nucleus segmentation algorithm based on multi-scale fuzzy clustering and a measurement method of interested nodes based on area prior are proposed.A multi-scale fuzzy clustering method is used to divide the cell cluster region,and a hierarchical tree structure is constructed for the cell cluster according to the inclusion relationship between the cell cluster division regions.The node of interest measurement method is used to select candidate nuclei from the hierarchical tree structure.The multi-scale fuzzy clustering method and the measurement method of interested nodes based on the area prior not only solve the problem of selecting the number of clustering categories of the clustering algorithm,but also improve the nuclear recognition performance.Regarding the problem of cell overlap and boundary blurring in cervical cell mass,a segmentation algorithm for overlapping cervical cells based on the enhanced radial boundary of cell nucleus is proposed.The proposed radial area filtering method can not only suppress the noise of cervical cell images,but also preserve the contrast of overlapping cell boundaries in the cervical image.The weight map generated based on the candidate contour points and contour line segment attributes is used,and a dynamic programming algorithm is used to find the shortest path in the graph.The level set model is used to finely segment the obtained coarse cell segmentation boundary,so as to obtain the final cervical cell boundary.In order to make better use of the depth information of the stacked focal planes in cervical cell samples,a nucleus and cytoplasm segmentation algorithm based on depth information is proposed.By including the circular features generated by the clustering of nuclei with depth information,this paper constructs a tree domain structure to find candidate nuclei regions in cervical cell images.In order to obtain a more accurate cell nucleus area,this paper uses the adaptive radius morphological expansion level set method to refine the segmentation of candidate nuclei.By adding the depth information to the contour points and the line segment attributes,a coarse cell segmentation boundary closer to the real cervical cell boundary is obtained.