Classification And Recognition Of Breast Images Based On Improved Convolutional Neural Network

The incidence of breast cancer ranks first among women worldwide,posing a serious threat to women’s life and health.Mammography,which is commonly used for screening,is considered to be the most effective means of diagnosing breast cancer.Currently,computer-assisted breast mass systems based on mammography can help doctors improve film reading efficiency,but improving the accuracy of assisted diagnostic systems and reducing the false positive rate are still challenging tasks.In the image classification field,convolutional neural networks have obvious advantages over other classification algorithms.In order to solve the shortcomings of long time and low accuracy in the diagnosis of breast lesions in the computer-aided diagnosis system,this study achieved the classification of mammogram images by improving the existing convolutional neural network,integrating various model features and introducing attention mechanisms.The contributions of this paper are as follows:(1)An improved VGG16 network image classification method for breast diseases is proposed.In view of the very small proportion of breast lesions in mammogram images,this paper made improvements by simplifying the number of convolutional layers and the number of convolutional nuclei in the VGG16 network model,replacing the fourth common convolutional layer with a deeply separable convolutional layer,and introducing ECANet attention mechanism behind the fifth convolutional layer.The experimental results showed that the accuracy of this method was 99.8% and 98.05% in the MIAS(Mammographic Image Analysis Society)and DDSM(The Digital Database for Screening Mammography)data sets,respectively,which was significantly better than some methods in recent studies.(2)A GoogLeNet-BC network model for breast disease image classification is proposed.In order to take advantage of the high convergence rate of the Goog Le Net model in the classification prediction of mammogram images,the Inception-S module is established by replacing the 5×5 convolution of the third group in the Inception V1 module with 1×1convolution,and the number of Inception-S is reduced.The attention mechanism of ECANet is introduced and the branch of auxiliary classifier is deleted.Although this model has overfitting phenomenon in the prediction of MIAS,it can indeed significantly improve the prediction efficiency without reducing the convergence speed of Goog Le Net model under the condition of sufficient sample size.(3)A classification method of breast diseases based on feature fusion is proposed.The features extracted from the above improved VGG16 and the improved Goog Le Net-BC network were integrated,and then ECANet attention mechanism was introduced,and the global average pooling layer was used to replace the flattening layer and the average pooling layer,and then the classification operation was carried out.The classification effect of this method was further improved in MIAS and DDSM data sets.In conclusion,the improved convolutional neural network model proposed in this paper is indeed relatively simple,and can achieve a certain balance between prediction accuracy and prediction efficiency,which has certain positive significance for the classification of mammography images.