Research On Cell Segmentation And Counting Based On Deep Learnin

By studying the distribution and number of cell locations in pathological images,we can better understand the characteristics and mechanisms of diseases,which helps pathological analysis and disease diagnosis.However,existing methods are unable to perform accurate segmentation and counting of complex cell images and have the risk of missing detection,so they are not well suited to practical clinical needs.In recent years,with the development of deep learning technology,many researchers have successfully applied its powerful feature extraction and expression capabilities to the automatic segmentation and counting of cells,thus greatly improving the diagnostic efficiency and accuracy.In this paper,we focus on the automatic segmentation and counting of complex cell images and implement a cell counting system as follows:(1)An improved network model MSE-UNet(Multiscale skip connectionsqueeze and excitation-UNet)is proposed for the segmentation difficulties such as blurred cell image boundaries and much noise.Based on the U-Net network structure,a multiscale skip connection suitable for complex cell images is designed,and a channel attention module is introduced to focus on the key features of the image to be segmented,effectively removing the cluttered background and obtaining clearer cell boundaries.The multiscale skip connection cleverly combines different levels of information together to avoid redundancy,segment the target more precisely and solve the problem of unclear cell boundaries.The channel attention module learns the weight information of each channel,so that the important feature channels occupy a greater weight.Experimental results on Dsb＿2018,Isbi and the self-built dataset Cell＿142 show that the model has better performance compared to other advanced cell segmentation models.The important index of Dice coefficient reaches0.9112/0.9522/0.9325,respectively.(2)To address the difficulties in counting cells caused by dense,overlapping and blurred boundaries,a two-stage cell segmentation method combining MSE-UNet segmentation model and watershed algorithm is proposed.Firstly,the MSE-UNet segmentation model is used for primary segmentation,and then the watershed is used again for fine segmentation of adherent and overlapping cells to obtain more accurate cell segmentation images,based on which the connected domain algorithm is used for cell counting to improve the counting accuracy.The performance test and comparative analysis of the two methods using self-built dataset verified that the proposed two-stage segmentation method can segment cells and perform counting more accurately and obtain better performance compared with the traditional counting method,with a counting accuracy of 97.53%.In addition,a medical-assisted counting system for cell images is designed and implemented to demonstrate the experimental results of cell segmentation and cell counting and to verify the effectiveness and practicality of the proposed method.