Research On Segmentation Algorithm Of Cell Microscopic Image And Its Applications

Cell,as the fundamental unit of life,has been studied since 17^th century,and people tried to understand its basic composition and function.So far,scientists have made numerous achievements on the study of cell functions and behavior,which greatly promoted the progress of clinical medicine and pharmacology.With advances in microscopy,biologists have been able to easily access a large number of high-quality images of cells,which also provides the basis for cellular analysis.Due to the richness of cell images,the manual management and analysis of cell images cannot meet the needs of the development of cell biology.Therefore,the development of efficient cell analysis algorithms has become a hot research direction.Image segmentation is one of the most important basic fields in digital image processing.It is also a central issue of many researches and can provide a basis for subsequent image analysis and understanding.Cell segmentation,as a branch of the image segmentation algorithm,aims to extract cells from the image and is the basis of subsequent cell analysis,is central to such tasks as cell tracking and counting.In the past few decades,a large number of cell segmentation algorithms have been proposed,such as thresholding,region growing,curve evolution,graph theory-based segmentation and image segmentation based on deep learning.As an image segmentation algorithm based on curve evolution,level set method is widely used in cell segmentation.The existing level set models are mainly divided into edge-based level set model and region-based level set model.In this paper,we proposed a level set model based on region-edge information:SDREL,which adds salience information into the energy function.We also propose a two-stage evolutionary method.We compared our SDREL with traditional level set method on natural image data set and cell images,and the results can verify its effectiveness.Cell tracking is a process of tracking cells in a microscopic image sequence.The input is a cell image sequence,and the output is the complete locus of each cell and the inheritance relationship between cells.Cell tracking is often used for cell analysis.In this paper,a cell tracking method based on cell segmentation is proposed.Firstly,cell segmentation is performed on the image using the level set method SDREL.In the tracking phase,the segmentation results are matched between frame and frame.Then,based on the distance information and prior knowledge,the matching results are corrected,and finally the shorter tracking trajectories are fused.In this paper,we test the tracking algorithm on three datasets,and the experimental results also verify its validity.In recent years,deep learning has achieved great success in the field of image,and also has achieved better results than traditional algorithms in image segmentation.In this paper,an adipose cell counting method based on deep learning is proposed.To solve the problem of insufficient dataset,an image warping method based on affine transformation is proposed to do data augmentation.And we use conditional adversarial networks to implement image transformation,from label image to cell image,which further expand the dataset.Using the expanded dataset,we trained a fully convolutional network,U-net,to do adipose cell segmentation and achieved good results.