Rapid Classification And Reconstruction Of Blood Cell Subclasses Phase Imaging Without Unwrapping

The morphological analysis,species identification and screening of diseased cells of blood cells are closely related to human health.Because most cells are transparent phase bodies,it is difficult to achieve clear imaging by using traditional microscopic imaging technology.Although fluorescence detection technology can achieve high-resolution imaging of cells,its fluorescent calibrators belong to chemicals,which may change the morphology of cells and increase the discrimination of cells.With the progress of science and technology,quantitative phase microscopy has become the main optical method for the study of cell morphology and biodynamics with the advantages of label free,non-invasive,high sensitivity and quantifiable.The research results of this technology are mainly divided into two types:interferometry and phase unwrapping.Therefore,from imaging to reconstruction will consume a lot of time,which is not conducive to the rapid classification and recognition of batch cells.Considering the spherical like characteristics of blood extracellular cells and the sensitivity of wrapping phase to the characteristics of cell subclasses,a rapid classification and recognition method of blood cell subclasses based on polar coordinate system and non-unwrapping phase diagram without morphological reconstruction is carried out in this paper,so as to achieve the technical goal of rapid classification and recognition of batch blood cell subclasses.The main research contents are as follows:Firstly,based on the spherical characteristics of blood extracellular class,this method transforms the cell wrapping phase diagram in rectangular coordinate system into wrapping phase diagram in polar coordinate system,and proposes to establish the characteristic quantity characterization system of wrapping phase diagram of cell subclass,which is: symmetry H,roundness P and singularity N.With the help of MATLAB simulation software,the automatic classification of three classifications of blood cells in small batches is realized.On the basis of good classification results of simulation experiments,the wrapped phase diagram in polar coordinate system after graphic processing is taken as the data set.Through the convolution neural network training based on the classical network of Lenet-5 and the random gradient descent algorithm in back propagation.Ultimately,the rapid recognition of three-classification of batch blood cells is realized.Simulation and experimental results show the feasibility of this method.Secondly,in order to further expand the classification and recognition range of recognizable blood cell subclasses,this paper proposes a new feature fusion algorithm,that is,a feature fusion algorithm combined with two branches of convolutional neural network to improve the accuracy of image attribute description,and it also discusses the three cases of global feature fusing HOG feature,LBP feature or both,respectively.The results show that the network can be successfully applied to the rapid recognition of five-classification of blood cells.Based on the phase microscopic imaging technology,in view of the slow speed of traditional phase imaging for batch cell morphology recognition,this paper proposes a new blood cell subclass classification and recognition method by using the characteristics of globular like morphology of blood cell exosomes and the sensitive reflection of wrapping phase on its substructure morphology,and the application of extended depth learning computing technology in this field.The simulation experiments and experimental verification show the effectiveness of this method.This can provide a new idea for the research and development of rapid recognition technology of batch blood cells subclass classification.