Study On Multi-View Mammograms Mass Performance And Detection Methods

The Computer-aided Diagnosis (CAD) system helps the declining of mortality. But in the nowadays, the technology of CAD system is not worked ideally. The sensibility of the CAD system is always not high to satisfy the clinical X-ray doctors and the positive rate is much high, it needs more development. In this paper, this paper study on the combining of the MLO and CC mammograms to detect the breast masses and the segmentation of the breast masses based on the dynamic programming method. It mainly contains four parts.(1) This paper study on the symptoms of breast cancer in mammograms. The breast masses are the most common symptoms and they are in varying size, shape, and density. Exhibit poor contrast between the breast masses and the breast texture density. Highly connected to the surrounding glandular tissue, particularly for speculated lesions. Surrounded by non-uniform tissue background with similar characteristics and so on.(2) This paper proposes an algorithm to detect breast masses based on multi-size Sech template. First, this paper abstracts the breast region from the mammograms, and makes the enhancement. Then this paper use two different size templates to identify and record suspicious pixels, combine, and grouped suspicious pixels with a dynamic threshold.(3) Multi-view mammogram masses detection methods includes the breast pectoral muscle detection in MLO-view mammograms, MLO-view and CC-view registration, determine the matching strips, calculate the distance features and region linking to matching the features. If the suspicious breast region not satisfies the requirement, this paper delete it, else this paper mark it.(4) In the mass segmentation processing, this paper use the dynamic programming method to detect the boundary of the mass. This paper change the ROI to the polar coordinate, calculate the local cost matrix, calculate accumulate matrix, and then find the optimal path through the back-tracing the path from the end pixel to one of the pixels the first column. The results conceived are acceptable. Through the study on the breast masses detection method based on the multi-view mammograms, with the database contains48sets of mammograms including48MLO mammograms and48CC mammograms. In the situation of100%sensitive in mass detection based on multi-size Sech template matching, this paper makes the false positive decline from87.06%to70.05%, and the false regions from7.14drop to5.75in every image. So the method proposed in this paper is effective, and it is significative for the CAD system.