Research On Karyotype Analysis Method And System Based On Human Chromosome Image

Chromosome karyotype analysis is the basic method of cytogenetics research.This method is mainly based on the metaphase image of chromosome in cell mitosis,according to the number and morphological characteristics of chromosomes,they are paired,sorted,numbered,and finally the karyotype map is obtained.In the field of biology and medicine,karyotype analysis is an indispensable means to understand the composition of biological cells,to study species evolution and to diagnose genetic diseases.However,the traditional process of human chromosomal karyotype analysis is tedious,which involves field of vision screening,image capturing,image preprocessing,target segmentation,classification and matching,and report generation,and the whole process is timeconsuming and laborious.In order to improve the automation level of chromosome karyotype analysis,this thesis focuses on key tasks such as preprocessing,segmentation and classification of chromosome images,studies and selects appropriate algorithms,implements existing algorithms and improves them.Specifically,it mainly includes the following aspects of work content:1.Image preprocessing: the image preprocessing is ready for the subsequent chromosome segmentation and classification,it mainly includes the comparison and selection of denoising algorithm and texture enhancement algorithm.2.Chromosome segmentation and classification based on traditional image processing: The existing segmentation algorithms are implemented and analyzed,and a segmentation scheme based on convex hull algorithm is proposed.The existing chromosome feature extraction methods are implemented and improved,and the classification performance of a variety of commonly used classifiers is compared through testing.Finally,the random forest is selected to classify chromosomes,and the classification accuracy of 0.65 is obtained.3.Chromosome segmentation and classification based on deep learning: with the support of a large number of real data,the instance segmentation network of Mask R-CNN is used to integrate the two tasks of chromosome segmentation and classification,so as to integrate the process of chromosome karyotype analysis.Furthermore,this thesis use the Mask Scoring R-CNN network to improve the existing problems of Mask RCNN.Finally,we get the recall rate of 0.992 and the accuracy rate of 0.945 on the data set of this thesis.4.Development of chromosome karyotype analysis software: Integrating the algorithms in this thesis and developing chromosome karyotype analysis software according to the actual user requirements.The software will cooperate with the chromosome automatic scanner to improve the work efficiency,thus greatly improving the intelligence of chromosome karyotype analysis.Based on the above work,this thesis combines the image algorithm and software system effectively,realizes the integration and automation of chromosome karyotype analysis process,and achieves the expected goal.