Research On Computer-Aided Breast Masses Detection Algorithms

Breast cancer is one of the most prevalent tumor diseases among women. The key of its prevention and curing lies in early diagnosis. Mammography is the first choice to diagnose breast cancer. However, features of early breast cancer in mammograms are unconspicuous, even the experienced doctors may miss some subtle lesions. With the rapid progress of imaging and computer technology, computer aided diagnosis system of breast cancer based on mammograms can provide a consistent and reproducible'second opinion'to doctors, which can reduce false-negative rates and improve true-positive rates.Masses are the major indications of breast cancer in mammograms. In this paper, computer-aided breast masses detection algorithms are researched. Firstly, for the problem that mammograms have a large amount of data, and that detecting suspicious regions in them directly is very difficult and time-consuming, according to the characteristics of breast masses, a fast algorithm for searching suspicious regions is proposed. The comparisons between three filtered results of mammograms and the utilization of morphological open operation make that it can search suspicious regions. Furthermore, in order to obtain the contours of suspicious regions more accurately, an image enhancement method based on the morphological analysis is improved, which not only can effectively suppress the background and enhance the lesions in mammograms, but also can filter black holes. On this basis, a segmentation algorithm of suspicious regions based on OTSU is proposed. It takes the searched regions as"seed points", and obtains initial rectangular regions of interest (ROI) from the enhanced images automatically. Then through modifying the threshold of OTSU and adjusting the borders of the ROI automatically, it can ensure that the whole suspicious regions are located in the ROI entirely and are separated perfectly at last. Finally, in order to reduce a large number of false-positive regions, some features are extracted from the segmented regions, and then classification is carried out.The experimental results demonstrate that the proposed detection algorithms are fast and effective. They can be applied to other tumors detection problems because of their generality.