Medial Axis Thinning Algorithm And Classification Model For Human Karyotyping

Human chromosome karyotyping plays an important role in examining genetic diseases,prenatal diagnosis and cancer pathology research.Chromosome karyotyping uses microscopic imaging equipment to capture images of metaphase chromosomes in cells,and then image processing and pattern recognition and other methods will be applied to identify each chromosome's label and sort all chromosomes to obtain a karyotype.The extraction of chromosome features and the selection of classification algorithms are extremely important to the entire process.The length,centromere index and profile features of chromosomes that are commonly used for classification are very dependent on the determination of the medial axis of the chromosome image.However,in the process of chromosome medial axis extraction,the medial axis extracted by the traditional thinning algorithm has the problems of easily forming more burrs,poor connectivity,and not being able to have a single pixel width;in addition,the classification accuracy and model's training time of the currently used chromosome classification methods still has room for improvement.Therefore,how to effectively extract the chromosome medial axis with less glitches,good connectivity,and single pixel width,and propose a classification model with better performance are the focus of this thesis.In order to extract the chromosome medial axis with fewer glitches,good connectivity,single pixel width and other characteristics,this thesis proposes a hybrid thinning algorithm based on sequential thinning and parallel thinning.The algorithm is divided into three stages: firstly,determining the chromosome's contour according to the edge detection algorithm and process the boundary noise;secondly,deleting and retaining the contour pixels according to the constraints;finally,post processing the medial axis obtained by the multiple iterations of the first two stages to ensure the medial axis is one-pixel wide.At the same time,in order to effectively evaluate the performance of the proposed algorithm,some mathematical-based evaluation index is proposed for evaluating the algorithm.By comparing the proposed hybrid thinning algorithm with other common thinning algorithms on existing chromosome datasets and public datasets,it can be seen that the algorithm proposed in this thesis can effectively extract less glitches and good connectivity,single-pixel wide features,and the algorithm has good generalization performance.Chromosome features such as the length,area,centromere index,and banding features are extracted based on the obtainning medial axis.In order to handle the shortcomings of the existing chromosome classification methods,this thesis proposes a competition queue model based on binary kernel extreme learning machine classifiers optimized by improved artificial bee colony algorithm competition queue model.The optimal penalty coefficient and kernel function parameters of each binary classifier for two types of chromosome classification are searched by improved artificial bee colony algorithm,and a correction program is added to the model.Through experimental comparison between the classification method in this thesis and the classification methods in related literature,it can be concluded that the classification method proposed in this thesis can effectively improve the classification accuracy and reduce the training time of the model.