Research On Cell Nuclei Segmentation Based On Deep Learning

Analyzing the images of cells and tissues is an important research field because it forms the basis for a large number of biomedical applications.Many complexities such as heterogeneous shapes of cells,overlapping cells,background noise,and changes in staining and lighting conditions make the automated analysis of microscopic images a difficult problem.Cell segmentation is the basic and necessary step in the analysis of microscopic cell images.Although many methods have been proposed over the years,the existing method have not been flexible and accurate.Therefore,it is necessary to propose a more accurate automated cell segmentation method.Semantic segmentation is one of the key problems in the field of computer vision.A range of applications,including autopilot,image search engine,virtual reality,and augmented reality,can greatly improve the application performance from the research progress of semantic segmentation.Such problem has been addressed in the past using various traditional computer vision and machine learning techniques.Deep learning revolution has turned the tables so that many computer vision problems are being tackled using deep architectures.Semantic segmentation is no exception.The adoption of CNN(Convolutional Neural Network)greatly enhances the effectiveness of various semantic segmentation algorithms.The main work of this article is reflected in the following aspects:(1)Modify the popular semantic segmentation algorithm and applied the modified algorithm to the cell nuclear segmentation.(2)This article improves on the common encoder-decoder architecture in semantic segmentation models.In the encoder stage,this paper add dilated convolution into the feature extraction network to solve the conflict between the two goals of extending the receptive field and maintaining the resolution of the feature map.Between the encoder and the decoder,in order to solve the problem of prediction of multi-scale objects,this paper designs a feature aggregation module that aggregates the output of different scales of the dilated convolution layer and the different scales of the pooling layer.To solve the difficulty of semantic segmentation model capturing multi-scale context information.In the decoder stage,a decoder structure with a residual link was designed to improve the effect.(3)For the cell nucleus segmentation task,this paper adopts data augmentation technology to expand the data set,and adopts traditional image processing techniques(contrast enhancement,morphological operations such as erosion,dilation,open operation,closed operation,etc.)to process the data set and results,and K-fold cross-validation is used during training to prevent overfitting of the model.(4)Detailed experiments were performing on the deep segmentation algorithm proposed in this paper to determine the contribution of each improvement to the overall improvement of the segmentation result.