Research On Breast Cancer Image Classification Method Based On Convolutional Neural Network

Nowadays,cancer seriously endangers the live and health of Chinese people,among which breast cancer ranks the first in the incidence of female cancer.Early diagnosis of breast cancer is of great significance for patients to receive timely treatment.Therefore,how to improve the accuracy of breast cancer classification and recognition has become an urgent problem to be solved.With the wide application and remarkable results of deep learning in the field of image recognition and classification,it has laid the foundation for deep learning in the classification of breast cancer pathological images,which has prompted many scholars to conduct a lot of research in this area.But there are still shortcomings.Therefore,this paper takes the Break His datasets as the research object,and uses multi-model fusion to achieve multiple classification under different magnifications.The main research contents are as follows:This paper proposes a method for classification of breast cancer pathological images based on convolutional neural networks.This method includes strategies such as data enhancement,transfer learning,model fusion and feature selection.Firstly,perform normalization and equalization preprocessing operations on the original image,and then use data enhancement techniques such as image rotation to expand the data set to solve the small problem of the data set;use the weighted cross entropy loss function Cross Entropy Loss to solve the sample imbalance problem.Secondly,selecting the CNN network(Alex Net,VGGNet16,VGGNet19,Inception＿v3,Res Net50,Dense Net161)pre-trained on the large-scale Image Net to train the classifier,and then select the optimal model Res Net50 and Inception＿v3 through the validation set loss rate,and then pass the concat Way to merge the two models.Then,In order to solve the problems of over-fitting and poor classification accuracy after feature fusion,the SDRR feature selection method is proposed.Finally,the fusion network is fine-tuned to achieve eight classifications of breast cancer pathological images at different magnifications.The method in this paper is verified on the Break His datasets,and the classification accuracy rate at the patient level and image level is as high as 94.18% and 94.12%,which is better than a single network,traditional machine learning methods and existing deep learning binary classification methods.It is suggested that the network is helpful to the classification of breast cancer pathological images and improve the efficiency of clinical medical diagnosis.