Research On Key Technologies Of Automatic Classification For Breast Tumors

The World Health Organization reports that breast cancer is one of the leading causes of death among women.In the past decades,the incidence of breast cancer in women around the world has been on the rise.Only timely screening for breast cancer,effective detection and diagnosis of breast cancer in the early stage can reduce the mortality of breast cancer.Breast cancer is regarded as a complex heterogeneous disease which occurs with a variety of clinical manifestations.Breast ultrasound,Magnetic Resonance Imaging,CT and X-ray are the common methods of breast examination at present.Mammography is the preferred method for early diagnosis of breast cancer,which has the advantages of simple operation and low cost,and has been widely used all over the world.Then the number of mammograms that need to be screened is enormous which brings a heavy workload to doctors.Computer-Aided Diagnosis(CAD)system based on mammography analyzes the mammography with various image processing methods,which can not only provide advice for doctors in clinical diagnosis and reduce doctors’ workload,but also improve the accuracy of early diagnosis of breast cancer.The masses and calcifications are the most common abnormalities in mammography and the main recognition objects of breast CAD system.More than 80% of breast cancer patients are diagnosed with breast masses as the first diagnosis.Therefore,the research object of this paper is masses.segmentation,feature extraction and classification of masses are the main components of breast masses CAD system.This paper mainly focuses on the important and difficult issues in the three main parts of CAD system,and establishes a complete framework of CAD for breast masses in order to improve the accuracy and efficiency of CAD.The main research results and conclusions of this paper are as follows:(1)The FCM algorithm based on PSO and GA is designed.In the classical FCM algorithm,the initial clustering center setting with randomness will affect the segmentation effect of the image.In order to solve this problem,this paper proposes a hybrid optimization segmentation algorithm of PSO and GA for FCM.The "memory" function of PSO algorithm can retain the best solution of the previous generation,and the crossover mutation and other operations in GA can produce high-quality optimization solution.The combination of them can not only solve the problem of premature maturity in GA-FCM and PSO-FCM,but also improve the stability of the algorithm.The segmentation results of mammography mass image from DDSM database show that the two validity evaluation indexes of(1 and(1 are better than GA-FCM,PSO-FCM,FCM,PCNN and K-means.The proposed algorithm can segment the breast mass region better.(2)The multi-scale fusion features of the breast mass region are extracted.Feature extraction of breast mass region is a key step in breast mass CAD system.In this paper,16 feature values of the geometric feature,gray feature and texture feature of breast mass region are extracted.The analysis of the feature values extracted from the breast X-ray mass images in CBIS-DDSM database shows that the feature values extracted in this paper have obvious distinctiveness.In view of the limitation of single scale image features,the multi-scale feature of 16×4 dimension of breast mass region is extracted based on Gaussian pyramid,and PCA algorithm is used to reduce the feature dimensions.The multi-scale fusion feature obtained can extract the feature information on the deeper structure of the mass region,which can improve the classification effect and the performance of CAD system.(3)The SVM classifier based on PSO parameter optimization is designed on the basis of the classical SVM.Aiming at the non-linear separability in the classification of benign and malignant breast masses,the RBF function is selected as the kernel function,and the eigenvectors in the original space are mapped to the high-dimensional space.The optimal classification hyperplane is found to realize linear separability while reducing the computation amount.The PSO algorithm is used to optimize the penalty parameter of the classical SVM and parameter γ of the RBF kernel function.The benign and malignant classification experiment is carried out using breast mass images from CBIS-DDSM database.The accuracy of the classifier designed in this paper is 92.85%,and the area value under ROC curve reaches 0.9269.The accuracy of classification is higher than the classical SVM classifier and other classification methods.It proves that the effectiveness of the multi-scale fusion features extracted and the performance superiority of the classifier designed in this paper,both of them can play a great role in breast mass CAD system.