Research On Key Technology For Computer Aided Diagnosis Of Breast Cancer With Mammograhpy

Breast cancer has been the primary disease which badly threatens women's health,during past decade,the incidence of breast cancer increases year by year and presents young trend,it is find that early detection and timely treatment can reduce the breast cancer mortality effectively.Mammography is a sensitive and specific method for early detection of breast lesions,the mammograms increase heavily which lead to a heavy burden on experienced radiologists,as well as the diagnosis may be affected by subjective factors.Therefore,using computer technology to locate the lesions and classify seems very important.While noise and low contrast of mammography X-ray images may lead to detection difficult.Meanwhile,the irregular shape of calcifications and masses also reduce the accuracy of computer-aided diagnosis(CAD).Hence,mammograms preprocessing,lesion detection and classification have been a hot research topic in the field of breast cancer CAD.This thesis mainly research on solving the key and difficult problems of mammograms processing and analysis,aim at improving the accuracy and effectiveness of CAD breast cancer,finally establish a complete CAD system.The main contributions and innovations are as follows:1.In order to reduce the noise in mammograms which may affect the lesions detection and classification,a new mammograms denoising method is proposed which combines with multiresolution and bivariate threshold function.Firstly,construct a new multiresolution transform named dual contourlet transform,which using the special structure of both contourlet transform and dual tree complex wavelet transform;secondly,establish the statistical properties based on dual contourlet coefficients and combine Bayes estimate theory to obtaining the corresponding bivariate threshold function;lastly,test this proposed method on nature image,and then apply this method on mammograms denoising.The experimental results show that this proposed method can remove noise and retain the mammograms structure and texture information simultaneously,moreover it can obtain higher peak signal noise ratio and structure similarity.2.To improve the accuracy of breast cancer diagnosis,a new automatic mass segmentation and classification method of mammograms is proposed.Mammograms always face with low contrast and blurred mass edge,this paper puts forward a new method to solve this problem and segment the mass,which combines with rough set theory,Hough transform and improved VFC snake model;secondly,extract feature based on the segmented mass,Region of Interest(ROI)and corresponding background,and then the feature database is established;finally,the random forest is employed for classification between benign and malignant,and the state of the art methods are used for comparison,which include Support Vector Machine(SVM),Genetic Algorithm-Support Vector Machine(GASVM),Particle Swarm Optimization-Support Vector Machine(PSO-SVM)and decision tree.This proposed method is tested on classical and common database-DDSM database and MIAS database,the experimental results show that this proposed method can detect the mass accurately and obtain higher diagnosis accuracy between benign and malignant.In clinic,the doctor could treat the disease timely and accurately according to the results of computer diagnosis and real situation.3.The accuracy of segmentation will affect the diagnosis accuracy,to solve this problem,a new classification method based on the ROI is proposed.The proposed method is inspired by other ROI-based classification references,extract the ROI manually according to the gold standard without segmentation;then,decompose the ROI with the new dual contourlet transform and extract the feature based on the transform coefficients,these features include gray,texture and other nine features;finally,the improved k nearest neighbor(KNN)is employed to classification.MIAS database is used for testing this proposed method.The experimental results show the availability of the proposed method,and it can be considered using for the system of breast cancer CAD.In addition,in order to remove the interference information in mammograms,an improved breast region extraction method is proposed,which based on both the mathematical morphology and maximum connected region.Firstly,adjust the gray-scale of mammograms and obtain binary image;secondly,remove the background region with calculating the maximum connected region and other basic operations;finally,mathematical morphology is used for extracting the breast edge.Experimental results prove the effectiveness of the proposed method.