Research On Methods Of Chromosome Automatic Segmentation And Extraction In Chromosome Micrograph

In medical examination,the chromosome image automatic analysis system is generally used to process and analyze the chromosome micrograph,and to assist the inspector in the diagnosis of related diseases.In view of the fact that the current some chromosome automatic analysis systems have a low degree of automation when segmenting and extracting chromosomes,this thesis takes chromosome micrograph as the research object,and designs a series of methods that can automatically segment and extract chromosomes from chromosome micrograph.The thesis focuses on solving two problems: one is how to completely extract each chromosome region from the micrograph automatically;the other is how to effectively segment overlapping and conglutination chromosomes automatically.The main research contents are as follows:First of all,when performing background segmentation on chromosome micrograph,considering that the Otsu method is used to determine the image binarization threshold will appear biased,this thesis proposes an improved threshold determination formula,which uses a weight parameter to consider both intra-class variance and inter-class variance differences of between background and chromosome region,and finally determines the optimal threshold of the micrograph.In addition,in order to improve the processing efficiency of the algorithm,this thesis proposes an annealing genetic algorithm to further optimize Otsu,which incorporates the principle of simulated annealing into the genetic algorithm,and determines the optimal threshold by optimizing within the gray range of the micrograph.The experimental results show that the method proposed can not only ensure the structural integrity of each chromosome after segmentation,but also improve the processing efficiency.Secondly,in order to prevent chromosome structure missing in automatic segmentation results,this thesis innovatively proposes a method of using neural network to segment the overlapping regions in the image before segmenting overlapping and conglutination chromosomes.Because the poor performance of segmentation networks,an improved U-Net is designed,which is based on the original U-Net and incorporates the advantages of residual units and batch normalization to segment the overlapping regions of chromosomes.In addition,this thesis uses existing chromosome data to automatically generate various chromosome images and label images that can meet the network training through operations such as rotation and translation,which solves the problem of less training data and the difficulty of labeling overlapping regions.The experimental results show that the improved U-Net segments the overlapping regions of chromosomes more completely and accurately,and the results are more conducive to the automatic processing of overlapping and conglutination chromosomes.Finally,when automatically segmenting overlapping and conglutination chromosomes,the images were divided into overlapped chromosomes and non-overlapped chromosomes by segmentation results of the improved U-Net.For the overlapping chromosomes,this thesis proposes the segmentation idea of "separating first and then splicing",the overlapping regions are removed from the image,and then different methods are designed to splice non-overlapping regions according to the quantity and geometric characteristics;For non-overlapping chromosomes,this thesis first proposes the convex hull of the target region duty ratio threshold method to judge the conglutination chromosome,considering some severely bent single chromosomes may be misjudged,so for the images that were initially judged as conglutination chromosomes,this thesis proposes a method to find whether there is a set of optimal segmentation pits in the chromosome region to further determine conglutination chromosomes,and eventually use the determined optimal segmentation pits to segment the conglutination chromosomes.The experimental results show that the automatic segmentation method proposed can segment overlapping and conglutination chromosomes more accurately and extract each chromosome in the micrograph more effectively,which has better practical value.