Comparative Research And Implementation Of Cell Counting Methods Based On Density Estimation And Based On Position

Cells are the basic functional unit of a living body.The measurement of cell counts can effectively assist the diagnosis and prevention of diseases in clinical medicine.Initially,manual cell counting was inefficient and subjective.The cell counting instrument that appeared in the middle of the 20th century has liberated manpower and has been evolving so far,but it’s far from enough in intelligence.In recent years,deep learning has made great breakthroughs in computer vision and other aspects with the help of supercomputing capabilities to handle complex learning tasks.Applying deep learning to microscope cell images to achieve cell counting tasks has also become a hot topic in biological research.This paper comparative studies two mainstream cell counting methods based on deep learning by their algorithm theory and model implementation,and finally develops application systems to implement algorithm functions.The project focuses on the algorithm theory and model implementation of cell counting methods based on density estimation and based on position.The cell counting method based on density estimation uses Gaussian kernel density estimation and histogram density estimation to obtain density maps respectively,and regress the density maps in order to achieve the cell counting by using the fully convolution neural network.The cell counting task based on position is separated into position task achieved by cell segmentation and counting task.First,the improved U-Net is used to achieve end-to-end generation of cell segmentation image.Second,VGG-11 is trained to calculate the number of cells in the segmented image.Based on that,valid conclusions are drawn from the comparison of the performance of these two methods in different data.In summary,the cell counting method based on density estimation is suitable for target counting which has high-density,small and fuzzy shapes,and the other method based on position has a higher accuracy rate in the case of obvious target shape and low density.Finally,software engineering method is used to design and implement a cell counting system based on density estimation and based on position.Users can use this system to count their own cell images.