Research Of Microcalcification Detection And Computer-Aided Diagnosis Techniques Based On Scale Space Filtering

Breast cancer is a common malignancy of women. Prevention is the key to early diagnosis and treatment. At present, one of the key technologies to achieve early diagnosis is to detect microcalcifications of mammograms and to determine whether they are malignant. Because clustered microcalcifications are important signs to early breast cancer. So the increase in the detection and classification of microcalcifications in mammograms has been of interest to many researchers. This paper proposed algorithms of the mammograms enhancement, the microcalcifications detection and the lesion type diagnosis based on analyzing the characteristics of microcalcifications.Firstly, in order to improve the badly vision of microcalcifications which were affected by noises and tissues, a novel mammogram enhancement algorithm based on shape-selective filter and adaptive background suppression is proposed. The supposed algorithm detected the potential non-linear microcalcifications using shape-selective filter, divided the mammograms into foreground and background regions, then processed the background information using adaptive contrast suppression method, and enhanced the foreground information at the same time. It could selectively enhance the key information of the mammograms.Secondly, In order to improve the problem of true-positive and false-positive in microcalcifications detection, a novel microcalcifications detection algorithm of mammograms based on multi-scale space filtering and l1 norm nearest-neighbor classifier is proposed. The salient feature images are obtained by using multi-scale space filtering for original images, and the coarse detected binary image of microcalcifications is obtained via using microcalcifications segmentation method based on human visual model. Then the effective features are extracted and sent into the l1 norm nearest-neighbor classifier to remove the false-positive.Finally, according to the differences in shape, size and distribution between the binign and malignant calcification, the characteristic information of microcalcification clusters are extracted based on the microcalcification detection and sent into the second-level l1 norm nearest-neighbor classifier to judge. It could accurately extract the microcalcification clusters and classify the lesion type.