Study Of Breast X-ray Image Enhancement And Classification Of Benign And Malignant Tumors

Breast cancer is a serious threat to women’s life and health and has become one of the most common malignant tumors among women in the world.In recent years,the morbidity and mortality of breast cancer have shown a rapid growth trend.The early detection and timely treatment of cancer will greatly reduce the mortality rate of patients,so the early detection and diagnosis of the disease will be particularly important.At present,there are many diagnostic methods for breast cancer.The most common detection method is mammography.However,due to the limitation of imaging conditions,the clarity of some images is not high,thus affecting the doctor’s diagnosis.Mass is one of the most common direct signs in breast X-ray photography images.Through the analysis of mass images,the benign and malignant of mass can be judged.Based on this background,the main contents of this paper are: enhancement of breast X-ray images and classification of benign and malignant masses.(1)The A new adaptive enhancement method of breast X-ray images based on Non-downsampled contourlet transform(NSCT)is proposed in this paper.First,the breast X-ray image of the breast is subjected to histogram equalization processing to enhance the overall visual effect of the image.Then,the histogram-equalized image is decomposed in NSCT domain,and its high frequency is decomposed into three layers,each layer is divided into 2,4 and 8 directions respectively.The first and second layers are median filtered to remove noise,and then a special edge filter is designed to enhance each sub-band coefficient,which involves two parameters.Whale Optimization Algorithm(WOA)is used to optimize the parameters and blind image quality index(BIQI)is used as the objective function of optimization.The filtered and enhanced sub-band coefficients are subjected to NSCT inverse transform to reconstruct the image,and finally the purpose of adaptive strong breast X-ray image is achieved,so that the details of the image are enhanced.Using the enhancement method proposed by data test in DDSM database,the enhanced image quality was evaluated objectively by using six evaluation indexes,namely information entropy,average gradient,standard deviation,contrast improvement index(CII),BIQI and comprehensive index,and compared with several current similar methods.Indicators show that the proposed method has a good enhancement effect on breast X-ray images.(2)Using a four-layer convolution neural network to classify the tumor images before and after enhancement into benign and malignant.The convolution neural network is used for autonomous learning of features.Four convolution layers,four pooling layers and two fully connected layers are used.The soft max layer is finally used as the classifier,the first 10 layers are used for feature extraction of images,and the last soft max layer is used for classification to realize the classification of benign and malignant breast X-ray images.The experiments on DDSM database show that the accuracy of image classification before and after enhancement is 92.25 % and 95.38 % respectively.AUC values were 0.96 and 0.97,respectively.The index shows that the enhancement method proposed in this paper has a certain effect on the image in the four-layer convolution neural network,and compares it with the classification method of similar images,showing that the classification method adopted in this paper has certain advantages.(3)The Residual Network is used to classify benign and malignant tumor images before and after enhancement.The deeper the convolutional neural network layer is,the more detailed the image feature extraction is,and the better the classification effect is.However,the gradient explosion will occur if there are too many layers,which will make the classification effect worse.Therefore,this paper adopts a 50-layer residual network(Resnet-50),which has a deeper layer and can extract rich and differentiated features.In-depth supervision of the framework based on ResNet-50 network can further improve the classification performance by directly guiding the training of the upper and lower layers of the network.DDSM database is used to verify the performance of the method.The accuracy of image classification before and after enhancement is 93.35% and 97.96% respectively,and the AUC values of images before and after enhancement are 0.96 and 0.98 respectively.The experimental results show that the enhancement method proposed in this paper can improve the classification accuracy of breast X-ray images in the residual network.Compared with the classification effect of four-layer convolution neural network,the results show that ResNet-50 network has more advantages in classification effect.