Automatic Detection Of Micro-droplet From Fluorescence Microscopic Images

Microfluidic technology has made great progresses in single-cell analysis,digital PCR and high-throughput screening.In the development and application of droplet-based microfluidic technology,the qualitative and quantitative analysis of micro-droplet content plays an increasingly important role.With this technology,the detection of quantitative biological characteristics such as the concentration of tiny biomolecules can be converted into indirect quantitative analysis of micro-droplets.Quantitative analysis of microdroplets often requires fluorescence microscopy imaging techniques and analytical processing of micro-droplet fluorescence images.However,the counting,detection,and analysis of micro-droplet fluorescence images mostly rely on the time-consuming reading of fluorescent images by biological researchers,which always requires a large amount of work and greatly increases researchers' burden.In order to solve the above problems,automated methods for detecting fluorescent targets have received extensive research and attention.However,due to the weak luminance difference between target micro-droplets and the backgrounds in the fluorescence image,complex noise environment,and the emergence of a large number of aggregation adhesions,there is still no effective automatic detection method for micro-droplets.Aiming at this problem,this thesis takes a deep study in this question and proposed effective automatic detection algorithms of micro-droplet in fluorescence microscopy images.According to whether the background of the micro-droplet fluorescence microscopy image contains background fluorescence induced by free enzymes,the micro-droplet fluorescence microscopic image can be divided into the micro-droplet fluorescence image with background fluorescence and the micro-droplet fluorescence image without background fluorescence.For the micro-droplet fluorescence image with background fluorescence,firstly,we apply Gaussian denoising filter to filter out the noise of the approximate Gaussian distribution in the image,and then use the contrast-limited adaptive histogram equalization(CLAHE)algorithm to globally enhance the image to improve the contrast and visibility of the empty micro-droplets.Then,the Otsu threshold segmentation which can effectively separate the foreground target and the background is used to obtain the binary segmentation map of the micro-droplet.Finally,the modified Hough circular transform algorithm capable of detecting the circular feature target is adopted to automatically extract and count all the micro-droplets in the image,and thus completing the automatic detection of micro-droplets under this type of image.In contrast,the micro-droplet fluorescence image without background fluorescence has lower visibility,the nonluminous empty droplets have poor contrast with the complex backgrounds and noise interference is great.Therefore,it is difficult to recognize the micro-droplets and even the state-of-the-arts fluorescence target detection methods are unable to detect these micro-droplets.This thesis designs a more effective algorithm for automatic detection for micro-droplets in this type images by analyzing the characteristics of micro-droplets in such image: firstly,through Anscombe variance stability transformation converting the Poisson-Gaussian mixed noise contained in the image into a Gaussian noise type,then using the newly developed adaptive clustering and progressive PCA approximation combined noise reduction algorithm to carry out the detail preserving denoising,then adopting the low-brightness image enhancement method based on luminance map estimation to adaptively enhance the local contrast of the empty micro-droplets falling in the dark background of the image after noise reduction.Furtherly,Radon-like feature extraction and adaptive threshold segmentation are selected for target micro-droplet binarization by exploring the edge feature of micro-droplets.Finally,the modified Hough circular transform algorithm is used to automatically detect the luminous fluorescence micro-droplet and the nonluminous empty micro-droplets,ensuring the feasibility and accuracy of clinical biomedical applications.Furthermore,these two proposed micro-droplet detection algorithms were tested for two groups of 15 fluorescence micro-droplet images respectively.The mean and standard deviation of true positive rate(TPR)and false detection rate(FPR*)and F-measure obtained by the detection algorithm for fluorescence images with background fluorescence are 0.9980 ?0.0022,0.0015 ?0.0019 and 0.9983 ?0.0013,respectively;the mean and standard deviation of true positive rate(TPR)and false detection rate(FPR*)as well as F-measure obtained by the detection algorithm for fluorescence images without background fluorescence are 0.9944 ?0.0071,0.0306 ?0.0346 and 0.9824 ?0.0171,respectively.The comparative experimental results show that these two automatic micro-droplet detection algorithms can effectively detect and count the micro-droplets in the fluorescence microscopy image.Compared with the existing fluorescence spot detection algorithms,the proposed methods are more accurate and robust for detecting micro-droplet targets in fluorescent microscopy images,which can meet the needs of clinical diagnosis and biological experiments.