An Modified Dynamic Programming Algorithm For Mammography Breast Mass Segmentation And Diagnosis

Currently, breast cancer is a public health problem has attracted extensive attention in the world. It is a kind of malignant tumor always in women, leading high death rate. The incidence of breast cancer in the European and North American countries is at the highest level and the second is Asia. According to the survey, there is one woman in12women suffering from breast cancer. A research showed that, the early diagnosis of breast cancer has been proved the most efficient measures to improve the cure rate and reduce the mortality of breast cancer.Mammography is the best check technique for early diagnosis of breast cancer, mass is major performance in mammography. Every year, there are a lot of mammographic images to be diagnosised, and it is a challenge task for radiologists. We developed a computer aided diagnosis system to assist radiologists as the second references. However, breast mass segmentation is the most important step in a computer aided diagnosis system, and it is a challenging task due to the characteristics of the image itself.This paper proposed an improved segmentation method based on dynamic programming and combined it template matching for segmentation and recognition of the breast cancer in mammography. This hybrid segmentation method can achieve effective and accurate segmentation of breast masses. Specific contents in this paper was arranged as follows:First, we roughly matched mammography image by the template matching method and located the tumor; then, we adopted an improved dynamic programming method to segment and depict the edge of the mass accurately; Finally, we utilized the maximum edge guidelines to extract more compact characteristic set from original characteristics set. Then, we used classifier to discriminate the benign and malignant, according to mass characteristics set. Our method can identify the mass edge more accurately and efficiently, it is advantage for the subsequent identification work. The effectiveness of proposed method has been proved on the DDSM database.