Methodology Study Of Single Cell's 3D Morphology And Diffraction Image Features For Classification

The traditional flow cytometry can only perform functional analysis of cell,it can extract the cell molecular information by qualitative and quantitative analysis of some specific components inside the cell.The fluorescence microscopic imaging flow cytometry has been developed in recent years for automatical and rapid analysis of cell images,but the cells have to be stained with fluorescent dyes in advance and it can only provide limited cell 2D information rather than the whole 3D morphological information.Our group has developed a novel diffraction imaging flow cytometry(DIFC)technique which can obtain the diffraction images of single cells' 3D morphology stimulated by coherent light and extract the diffraction image features which is highly correlated with cell morphology features.This dissertation aims to establish methods for quantitative analysis of the 3D morphologies of microspheres and cells,simulation of the diffraction images using reconstructed cell structures,analysis of diffraction images for 3D morphological feature extraction and classification of cells based on the diffraction image features.This dissertation tested and proved the feasibility of analyzing 3D morphology of particles using their diffraction images.Four samples of polystyrene microsphere with nominal diameter of 2.5?m,5.7?m,7.9?m and 9.6?m,respectively,were used to test the diffraction imaging experimental system and more than 1000 diffraction images were acquired for each sample of microsphere.An algorithm based on short time Fourier transform(STFT)was applied to obtain the local frequency information of microsphere diffraction images.Then a linear relationship was established between the local frequency information and microsphere diameter.It was shown that microsphere diameters could be obtained directly from the diffraction image frequency information.Furthermore the same algorithm was used to analyze the diffraction images of double microspheres which built up the foundation for future development of fast accurate microsphere measurement method.In cell study,a large number of cell diffraction images were obtained and through the observation of these images,it was found that different patterns were presented in the images.Through simulation calculation for cell and cell debris,three types of diffraction images were classified as cell,cell debris and noncellular particle diffraction images.According to the characteristics of diffraction images,a prescreening method was established for processing cell diffraction images and removing the diffraction images of cell debris and impurities automatically which should increase the cell classification accuracy based on diffraction images.Based on our previous work,the grey level co-occurrence matrix(GLCM)algorithm was chosen to analyze cell diffraction images for classification of cell lines and primary cells based on their 3D morphological features.In the experimental investigation,normal prostate cell(PCS),prostate tumor cell(PC3),Jurkat T cell(Jurkat),Ramos B cell(Ramos),mouse spleen T cell(T-cell)and B cell(B-cell)were studied.Features extracted from diffraction images were used for classification using support vector machine,and the classification accuracy of above 90% was achieved.In this dissertation a method was developed for precise simulation of diffraction images which is a very useful tool for correlation study of cell 3D morphology features with diffraction image features.First,the cell 3D morphology was reconstructed with their fluorescence confocal image stack.Then cell optical models were established according to the fluorescence intensities.And Mueller matrixes were calculated using discrete dipole approximation algorithm with different nucleus volume or different nucleus refractive indexes.The diffraction images were created by projecting the 3D images onto an imaging plane with Zemax software,and the images were then analyzed and classified with inhouse software tool of contourlet transform(CT).Finally the GLCM and CT algorithms were compared for analysis of diffraction images,and it was found that the GLCM algorithum is more suitable for the analysis and 3D morphological feature extracion from cell diffraction images.