Optimization Of Convolutional Neural Network For Breast Disease Diagnosis

Fast and accurate medical image classification is an important prerequisite for the application of computer-aided diagnosis technology in medical field.With the rapid development of medical imaging technology and the continuous improvement of computer performance,the use of computer-aided disease diagnosis has gradually become a research hotspot in the medical field.In recent years,with the rise and rapid development of in-depth learning,more and more scholars have applied it to the field of medical image processing.Among them,convolutional neural network,a typical deep learning model,has made some achievements in image classification because of its local connection and weight sharing.However,due to the limited learning ability of traditional convolutional neural network model and the complexity of medical image itself,its efficiency can not meet the practical application very well,so it needs to be done on the traditional convolutional neural network model.Some improvements are made in order to achieve the effect of practical application.This paper studies computer-aided breast disease diagnosis from three aspects: model improvement of convolutional neural network,feature fusion and classifier.The main contents in the research are as follows:A new diagnostic method for breast diseases based on improved convolutional neural network LeNet-5 is proposed.Aiming at the problems of low accuracy and long time-consuming of computer-aided diagnosis of breast diseases,a new diagnosis method of breast diseases based on improved Convolutional Neural Network(CNN)is proposed.This method has been improved in three aspects: firstly,a two-channel convolution neural network is designed to solve the problem of insufficient feature extraction in a single channel;Then,use dropout technology to effectively prevent over-fitting;finally,support vector machine(SVM)is used to replace the traditional softmax classifier to reduce the computational complexity and improve the total speed.After testing,the average accuracy of the proposed classification model is 92.31%,and the average training time is 968 s,which fully verifies the effectiveness of the proposed method.A convolutional neural network based on multi-scale feature fusion for breast disease diagnosis is proposed.In order to improve the accuracy of computer-aided breast disease detection,a breast disease detection algorithm based on convolution neural network is proposed.Firstly,the shallow and deep features of the image are extracted from the convolution neural network and weighted fusion is carried out.Secondly,the spatial pyramid pooling layer is constructed in the convolution neural network to realize the multi-scale input of the convolution neural network.Finally,the validation experiment is carried out on the data set of the Mapographic Image Analysis Society(MIAS).The experimental results show that the proposed breast disease detection algorithm'average accuracy is 94.93%.Compared with other breast disease detection algorithms,the proposed breast disease detection algorithm has higher detection accuracy.(3)A diagnostic method of breast diseases based on CNN multi-layer feature fusion and ELM is proposed.In order to solve the problem of low accuracy and long time-consuming of traditional computer-aided diagnosis methods,a method of breast disease diagnosis based on Convolutional Neural Networks(CNN)multi-layer feature fusion and Extreme Learning Machine(ELM)is proposed.First of all,use CNN to extracted multi-level features from mammograms using CNN;secondly,multi-scale pooling operation is proposed to fuse the features extracted from each layer;finally,the classification of mammograms using extreme learning machine is used for rapid diagnosis of breast diseases.The experiments shows that the average accuracy of the proposed breast disease detection algorithm is 97.13%,and the diagnostic time is about 6.43 Ms.This method can effectively improve the accuracy of breast disease diagnosis,shorten the diagnosis time,and has good robustness and generalization ability.