Preliminary Study On The Measurement Of Yeast-budding-rate By Diffraction Imaging Flow Cytometry

Aiming at the problems of slow,inefficient and poor stability of the traditional yeast sprouting rate calculation,this paper proposed a method to measure the sprouting rate of yeast based on the diffraction imaging flow cytometry(DIBM for short).The method comprises 5 steps: 1)Preparing a suspension of a sample of yeast to be tested,2)Performing a diffraction imaging flow cytometry measurement of the suspension,obtaining a large number of diffraction images of individual yeasts(budding or not budding)to a computer 3)Preprocessing and extracting gray-level co-occurrence matrix(GLCM)features of the diffraction images 4)Obtaining the classifier using machine learning to divide the yeast diffraction images into the budding group and the non-budding group according to the image features 5)Obtaining the yeast budding rate by counting the proportion of the budding group in the total yeast.The main tasks of this topic include the followings:1)Acquisition of training data and validation data for machine learning classifier,including: a)Culture the yeast to obtain a plurality of sprouting samples,b)Perform diffraction flow cytometry experiments to obtain a large number of single yeast diffractograms,preprocess to remove diffractograms which are under-dark and over-light;c)Divide the first group of yeast diffractograms into 1_training and 1_test two groups,artificial label 1_training yeast diffractograms based on prior knowledge to obtain training data;d)Take the control microscope for all groups of yeast,and artificially count the budding rates of the yeasts for the microscope to obtain the validation data which contains artificially counting the budding rates of each group's yeasts for the microscope,and yeasts' diffractograms which are not labeled;2)Gray level co-occurrence matrix feature extraction for diffraction image.The redundancy is removed by analyzing the correlation between the grayscale co-occurrence matrix features of the diffraction image,and the choice of the hyperparameters of the gray co-occurrence matrix is determined by visual analysis.3)Budding and non-budding yeast diffraction image classifier machine learning model establishment,training and effectiveness comparison.The machine learning method is selected by comparing the accuracy,stability and fastness of the classification model established by several machine learning methods,and also to verify the validity of the optimized gray level co-occurrence matrix features.4)The validity of the optimized DIBM method,repeated verification.The DIBM method was used to predict the germination rate of the six groups of yeasts and the correlation coefficient of artificial germination rate was 0.9759.The correlation coefficient of repeated experiments was 0.9625,which proves the validity and stability of DIBM method.At the same time,the maximum difference between the budding rate and artificial statistical results predicted by the model in repeated experiments was 1.7%.The results above preliminarily confirm the usability of the DIBM method and have a good demonstration effect on the analysis and classification of other cells,which is of great significance for the further application of the emerging diffraction imaging flow cytometry.