The Research Of Mammography Based On Convolutional Neural Network

According to relevant statistics from the World Health Organization(WHO),breast cancer is the most commonly diagnosed cancer among women in the world.Therefore,the diagnosis and treatment of breast cancer have attracted more and more attention.The golden standard for screening and diagnosis of breast cancer is mammography,which is also the most important technical method for breast cancer screening.Referring to the current research on computer-aided diagnosis systems for breast cancer,this paper proposes a classification method of breast density image and a benign/malignant classification model of breast tumor images based on the Convolutional Neural Network(CNN).Combined with deep learning technology,on the one hand,this network can learn identification features automatically to avoid designing specific hand-made image-based feature detectors;on the other hand,it provides a reliable basis for doctors to make the diagnosis and improves the diagnosis efficiency significantly.The main work in this article includes three aspects as follows.(1)The algorithm based on the U-Net is designed to segment the breast mass image and achieve accurate segmentation of the breast mass.The specific model is that after performing a series of preprocessing such as background removal on the breast X-ray image,the U-Net network is used to segment the mass in the breast X-ray image,and the information such as the segmentation contour of the breast mass region is accurately extracted.(2)A four-channel convolutional neural network is designed and the image of the mass lesion area is introduced as the fourth channel to train the convolutional neural network.The introduction of the lesion area will make the network pay more attention to the characteristics of the lesion area,and thus help doctors to make a more accurate diagnosis.The breast X-ray images labeled with the lesion area are trained,so that the network can converge quickly.Meanwhile,it solves the problem of falling into overfitting easily due to a small amount of data in the data set.Finally,accurate classification of the breast X-rays images based on a convolutional neural network is implemented.(3)A algorithm based on SE-Attention anti-aliasing mechanism networkis proposed to train breast density images and tumor images.Thereby,the network has the performance of feature recalibration and proceeds towards the direction of convergence quickly.This method can improve not only the classification consistency between different displacements,but also the accuracy of induction.Introducing anti-aliasing mechanism to convert and translate the model image,the proposed network can further overcome the overfitting caused by the scarcity of data and improve the classification accuracy and model generalization ability.Experimental verification shows that with the method proposed in this paper,the model can achieve the segmentation of breast X-ray lesions and the classification of breast images,which makes the model reach the current best state.The algorithm can provide a reference for the classification and diagnosis of breast tumors to a certain extent.