Study Of The Automatic Cell Image Segmentation Algorism Based On K-means Clustering And Mathematical Morphology

Pancreatic cancer is a malignant digestive tract tumor disease that is difficult totreat and diagnose. In recent years, the incidence of pancreatic cancer is on the rise. Tostudy and analyze the pancreatic cells has significance for the early prevention andclinical diagnosis of the pancreatic cancer. At present, the effective way to diagnose is toobserve the microscopic image of living cells slice and process, analyze the cell imageto get the diagnosis results. However, due to the staining, uneven illumination, thesampling methods and the other issues, in the images there are overlapping, clusters,adhesions cells and other interferences such as astigmatism. All these factors caused alot of difficulties for the diagnosis. To solve this problem we need to study an efficientsegmentation algorithm that can efficiently separate the overlapping, clusters, adhesionscells from the background. In the existing research, there are few public reports aboutthe pancreatic cell segmentation algorithm, therefore the problem need to be solvedurgently.For the multiple existing problems in the human pancreatic cell image, the thesisproposed an automatic segmentation algorithm that combined K-means clustering withadaptive morphology. The main contents are as follows:①By observing the characteristics of human pancreatic cell image, the thesisstudied and proposed a pre-segmentation method based on the K-means clusteringalgorithm. The method can cluster the background, interference and target prospects ofthe images and provide a locally connected domain for the subsequent segmentation.②Due to the fixed size and shape of the structure elements in the traditionalmathematical morphology algorithm, the thesis proposed a fine segmentation algorithmbased on the adaptive mathematical morphology. It can determine whether theconnected domain is adhesive or not and adaptively computing the size of structuralelements. Then conduct appropriate morphological operations, so the differentadhesions cells can be divided accordingly.③Based on the above two points, the thesis worked out an automatic pancreaticcell images segmentation method that is based on K-means clustering and adaptivemorphology. The method can also be applied to other tissue cell image segmentation.The segmentation results show that the algorithm can adaptively and accurately segmentthe overlap and adhesion cells. Compared with traditional morphological methods, the segmentation results in this paper are better.④Though the MATLB GUI software we designed a complete automatic cellsegmentation demonstration system that contains the classical image segmentationalgorithms and the algorithm in our thesis. The system can implement a variety of cells’segmentation and show the results. The thesis described the production methods of thesystem and used many pictures to show the running interface and the final segmentationresults.The results of the study showed that the method in this thesis can effectivelyseparate the overlapping and adhesive pancreatic cells from the complex imagebackground, the segmentation accuracy is better than some other cell segmentationalgorithms that are commonly used. In addition, the presentation system in the thesisshowed that the design structure of the system is simple, easy to operate and verypractical, by using the system we can implement a variety of cell image segmentationway.The research of this paper provided a novel theory and method for the automaticcell image segmentation and laid the foundation for health monitoring and auxiliarydiagnosis of the pancreas cancer.