Study On Texture Feature Extraction And Computer Aid Diagnosis Method Of Breast Mass

In recent years, the incidence of breast cancer shows a rising trend,but the mortality is decline. The reason is that mammography images can detect the breast masses detected impossible by doctors, which greatly improve the diagnostic accuracy of breast disease; to a certain extent, the breast Computer-Aided Diagnosis (CAD) system also can helps doctors identify the potential masses, which effectively improve the early diagnosis of breast cancer that has great significance for the cure of breast cancer.Breast mass is the most common symptoms of breast cancer, so accurate detection and location of the mass will greatly improve the precision of breast disease diagnosis. The mass organizational structure and surface roughness constitute its texture features, which is an important basis for distinguishing the mass. The paper mainly studies the CAD method of X-ray mammography and combines the texture features of mass, then an automatic texture feature extract method of breast mass is designed and finally detect the masses by support vector machine (SVM). The texture analysis and detection of masses part in X-ray mammography are achieved. Firstly preprocessing the mammography, according to the differences density of breast tissue and gray-scale layer to achieve the initial detection of suspicious areas; secondly extracting feature of initial inspection areas, according to the difference of breast tissue growth mechanism, the multi-level fractal features, Gray Level Co-occurrence Matrix (GLCM) and Tamura features are extracted, and the texture of breast tissue is achieved. Finally using the sort method to select these features and achieve the characteristics optimal; Last but not least, using the SVM algorithm to classify the suspicious areas and achieved diagnosis of the masses part.110 mammograms (including 115 cases of lesions) are diagnosis by above method, compared with the gold standard, and the method achieves a positive fraction of 93.04% with average of 1.1091 false positives per image. Experimental results demonstrate that the proposed detection algorithm is effective, and play an important role to improve performance of CAD system.