| Breast cancer is one of the most common malignancies in women.The nuclear lesion is one of the important signs of many diseases,especially cancer.The nuclei of cancer cells are obviously different from normal cells,so the characteristics of the nuclei are often used as an important basis for pathological diagnosis.The Nottingham histological scoring system was highly correlated with the shape and appearance of breast cancer nuclei in histopathological images.However,the complexity of automated nuclear detection is as follows :(1)large number of nuclear and highresolution digital pathological images are too small;(2)variability of the size,shape,appearance and texture of a single core.With the rapid development of medical image processing technology,computer-aided diagnosis system has gradually become an important method for doctors to diagnose patients.In recent years,people begin to pay attention to the application of "deep learning" strategy in the classification and analysis of large image data.Given its size and complexity,histopathology represents an excellent use case for the application of deep learning strategies.In this paper,a semisupervised antagonistic neural network(SGAN)is used to realize the automatic recognition of breast cancer nuclei,which is an example of deep learning strategy and can be used for fast,efficient and accurate automatic nuclear detection of highresolution breast cancer histophological images.Because the artificial labeling of medical histopathological images is an extremely tedious work,there is usually less labeling data.This article will focus on breast tissue pathological image data using a semi-supervised against neural network learning method,make full use of the mammary gland tissue pathology image training focused on a small amount with accurate data and a large number of labels without annotation data to realize automatic detection of breast cancer nuclei,combining unsupervised learning and supervised learning,network model for the disease to classify the course of study.Firstly,a nuclear detection method for breast cancer based on antagonistic neural network(GAN)is presented.The method consists of three parts: generating network,discriminant network and classifier.In the discriminator,VGG network was used to obtain the features of breast cancer nucleus,and the classifier Shared some features of discriminator to extract the network,the detection and recognition of the nucleus was realized by Softmax classifier.Then a semi-supervised network model which combines supervised learning with unsupervised learning based on adversative neural network is introduced,and the effectiveness and superiority of this model in detecting the nuclei of breast cancer histopathological images were proved by experiments.Experimental data were provided by the institute of pathology,affiliated hospital of Case Western Reserve University in the United States.The optimal classification accuracy of the model on this data was 94.1%,which was improved compared with other models.In order to verify the validity of the experiment,this paper the performance of the network model and the classic network model by comparing a variety of evaluation standard,proved a semi-supervised against neural network on the breast tissue pathological image performance is superior to the classic network,using a semisupervised against neural network is verified to examine the feasibility of breast tissue pathological image nuclei... |
