Nuclei Classification Of Pathological Images Based On Convolutional Neural Networks

In recent years,cancer has gradually become one of the main causes of death worldwide.According to the World Health Organization,cancer has risen from the second cause of death to the first cause of death in China.Diagnosing cancer at an early stage can effectively treat cancer,reduce cancer mortality,and increase the probability of cure.Therefore,accurate analysis and diagnosis of cancer are of great significance.It can not only improve the cure rate of patients but also reduce the cost of diagnosis and the burden of diagnosis for patients.The classification of cell nuclei is an important prerequisite for cancer diagnosis.Physicians analyze cancer or predict treatment through accurate nuclear classification.In recent years,some traditional methods based on image processing have achieved good results in the classification of nuclei,but there are still problems such as high requirements for professional knowledge.With the rapid development of deep learning,more and more scholars have begun to process images through convolutional neural networks,and have made great progress.However,most methods only classify single nuclei,and the classification of nuclei in histopathological images is still a great challenge due to the problems of cohesion,heterogeneity,and unbalanced distribution among the categories.In response to the above problems,this article studies how to classify cell nuclei efficiently,and achieves certain research results.The main work of this paper includes:1)Aiming at the problem that the background and the target in the medical image are relatively high in similarity,it is easy to misclassify the background and the target,and an improved cell nucleus classification algorithm based on the attention mechanism is proposed.This method is based on the convolutional neural network,uses the residual network structure and the feature attention module to perform feature extraction on the feature map,and uses three decoding branches for feature learning and restoration respectively.To better extract features,the attention module consists of channel attention module and spatial attention to obtain richer feature information.In addition,a subspace attention mechanism is applied at the end of the encoding module.The decoding branch is mainly composed of densely connected modules and convolutional layers.For the outputs of the three branches,this method uses a watershed algorithm for post-processing operations and then fuses the processed outputs to achieve the final nuclei classification.2)Aiming at the problems that the previous work only focuses on local features,lacks global features,and has a relatively large computational cost,a method for cell nucleus classification based on a parallel fusion of local features and global features is proposed.The method is based on a convolutional neural network,uses residual network structure and dense connection structure to extract feature map,uses multilayer perceptron after decoding module to learn global feature information,and uses multi-scale convolution module to learn More abundant local feature information,after the global feature information and local feature information are fused,a decoding module consisting of a convolutional layer and a dense connection module is used for feature learning and restoration,and three output branches are used to output the results.In order to To reduce the number of parameters of the network model,this method uses only one decoding branch for the three output branches.Likewise,after post-processing the outputs,the three outputs are finally fused to achieve the classification of nuclei.In summary,this paper proposes a specific solution for how to classify cell nuclei efficiently.The experimental results show that this method can realize the classification of cell nuclei well.Compared with the existing methods,this method not only achieves remarkable results in medical images with only a single tissue but also in many tissues and organs with the complex distribution.It also has better results in medical images.