Research On Segmentation And Classification Methods Based On Human Chromosome Images

As an important tool for relevant analysis and judgment in the medical field,medical imaging can play an important auxiliary role in diagnosis and treatment.Many intractable diseases that are difficult to be diagnosed by simply "watching,hearing and asking" can be traced through images,which is a great convenience Medical staff make accurate diagnosis and follow-up treatment based on the imaging results.The identification and analysis of chromosomal targets,as a great tool for scientific diagnosis and treatment of chromosomal symptoms at this stage,is the focus of current research and has a wide range of application values.However,the traditional process is completed by manual processing,which is cumbersome and inefficient.In response to actual needs and combined with actual image conditions,this experiment applies knowledge in the fields of digital image processing,pattern recognition,etc.,to study human chromosome image segmentation and classification methods,in order to obtain a more ideal chromosome automatic segmentation and classification system.This paper mainly contains the following experimental contents:Firstly,the thesis proposes an automated process that combines image filtering,contrast adjustment,morphological calculation,and detection and determination of the number of targets as criteria for impurity removal,and finally determines the chromosome region of interest.Approach.Secondly,the paper constructs an image edge detection algorithm based on shear wave transform,which has certain advantages compared with traditional edge detection algorithms through experimental verification.Then the image segmentation is performed to determine whether it is a single chromosome,and a watershed algorithm for segmentation of the hole-finding path and optimized threshold is proposed for the slight adhesion chromosome.The overlapping chromosomes are divided into single chromosomes by obtaining candidate pits on the edges of the subjects as dividing points according to their geometric characteristics.Thirdly,the paper formulates a feature extraction scheme based on the morphology of human chromosomes,and proposes a straightening method for curved chromosomes.Chromosome axis extraction is performed through operations such as chromosome refinement,pruning determination,and axis extension.The position of the centromere is determined based on the geometric characteristics and gray distribution.According to the chromosome morphology,the length and other characteristics are calculated.Fourthly,aiming at the actual situation of small samples and multi-classification,the paper analyzes and compares a variety of classification schemes through feature optimization to reduce the dimensionality of features,and determines that the random forest classification method used by the system shows the best classification performance.The experimental results show that the research on the segmentation and classification method of this paper has good results for chromosome recognition,achieves the expected goal,can improve the efficiency of processing problems in this field,and reduce the labor burden.