Masses Detection Algorithms For Mammograms

With the rapid progress of physic and Computer-aided Detection technology, people make more and more pressing demand for the early stage diagnosis of cancers. Mammography has been one of the most reliable methods for early detection of breast carcinomas. However, features of some early breast cancer in mammograms are inconspicuous, so it is difficult for doctor to discover all of the possible disease. A good computer aided detection (CAD) system can effectively avoid error and miss in diagnosis, the rate of the early stage diagnosis can be greatly raised. As the result, the patients will get more time for precautions and treatment and the incidence and death rate of this disease will drop considerably.Masses are the major indications of breast cancer on mammograms. For this purpose, this paper presents a new method to detect masses automatically on mammograms. Firstly, according to the characters of breast masses, a seed regions algorithm based on multi-layer ring filter is proposed, which extract plump seed regions use the gray and grads character of breast masses. Then a segmentation algorithm of suspicious regions of breast masses based on maximizing the between class variance (OTSU) is proposed, it can adjust the boundary of suspicious regions, which make sure the completeness of segmentation. Then a feature extraction method which calculates features from mass margins is proposed, in this paper, the mass boundary is modeled by a polygonal modeling, and then extracts a ribbon of pixels across mass margins through polygonal modeling, tumor and normal tissue is well separated by the features calculated from mass margins. Finally, the SVM classifier is designed to distinguish masses from normal areas. The experimental results conducted upon 100 cases show that the positive rate of our algorithm is 95.3% and got only average of 2.16 false regions per image, which demonstrate the validity and practicability of our method.The experimental results show that the proposed detection algorithms can obtain good detection accuracy. It is believed that these detection algorithms will be of extensive application prospect.