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

With the rapid development of society and the emergence of various medical instruments,there are more and more medical images in the real society.It is no exaggeration to say that the growth of medical images shows an exponential rise every day.In the face of so many massive medical images,people are also actively carrying out research,and strive for the national medical fields to make a contribution.Medical image classification has become an important research hotspot.However,due to the variety of medical images and the complexity of human body structure,the effect of traditional classification methods is not satisfactory.At the same time,people are also enthusiastic about the research of machine learning technology,and therefore,more advanced methods have been produced in the field of medical image classification.Among them,support vector machine(SVM)and deep learning network(DL)have been widely used in the field of medical image research because of their excellent classification performance.However,there are some problems in using machine learning to solve medical image classification:(1)First of all,the medical image itself has the following problems: high dimension,high complexity.High complexity makes it difficult to extract medical image features by traditional classification methods,and the extracted features do not have high discrimination.In addition,if the sizes of medical images are different,the network training will be affected,and the classification accuracy will be seriously affected.For example,when high-dimensional medical images are sent to SVM for classification,there will be the problem of dimension disaster.Therefore,in the research of medical image classification,it’s of great significance to find a suitable method to solve the problem of highdimensional image and extract effective medical image features.(2)The second is the problem of classification performance.Compared the traditional classification method,with machine learning method for classification,the classification effect has been greatly improved.However,as far as the machine learning method itself is concerned,its implementation process can be further optimized so that the classification effect can be further improved.For example,SVM is very dependent on the selection of kernel function and kernel parameters in medical image classification,and the optimization of kernel parameters of SVM can achieve higher accuracy in classification.Therefore,medical image classification has become an challenging problem.In view of the problems existing in most traditional classification methods: it’s difficult to extract medical image features,and the accuracy of the classifier needs to be improved.This paper proposes the idea of model fusion,and designs a high-precision medical image classification method based on deep learning,namely hybrid CQ-SVM.The research content of this paper is mainly divided into the following aspects:(1)Because of its powerful learning ability,deep learning method can effectively extract image features when it is used in medical image classification.Therefore,this paper trains a suitable convolutional neural network(CNN)for feature extraction of medical images.In our work,training a better CNN network is the focus of this part.At the same time,the trained CNN can also reduce the medical images dimension processing.The effectiveness of the method is illustrated by the experimental analysis.(2)In this paper,we choose SVM with excellent performance as our classifier.According to the previous research experience,we select the most commonly used RBF kernel function.In addition,we regard the kernel parameter selection of SVM as an optimization problem in the model.Quantum behaved particle swarm optimization(QPSO)is used to optimize the kernel parameters of SVM automatically.The experimental results show that the optimized kernel parameters make SVM classification performance improved further.(3)Based on the idea of model fusion,we regard the trained CNN as a trainable feature extractor,and the QPSO optimized SVM as a trainable classifier.The two are fused to form a new hybrid model,namely CQ-SVM model.This method can automatically extract features from original medical images and generate predictions.Experimental results show that this method can extract medical image features better and achieve high classification accuracy.Finally,we choose the breast histopathological images from the website of kaggle to do experiment and analyze the classification performance of CQ-SVM model,and make a comparative experiment on two aspects,one is the comparative experiment of medical image feature extraction,the other is the comparative experiment of classifier and optimization method.The experimental results show that,compared with the original model,the fusion model of the two structures achieves better results in IDC positive and negative classification of breast cancer pathological tissue images.