Automated Detection Of Microcalcifications

Breast cancer is the malignant tumor that threatens women's health seriously and the first cause of cancer-related mortality of women. Primary prevention seems impossible since the causes of this disease still remain unknown. Early detection is the key to improving breast cancer prognosis. Mammography is one of the reliable methods for early detection of breast carcinomas. An early sign of 30–50% of breast cancer detected mammographically is the appearance of clusters of microcalcifications(MCCS), and 60–80% of breast carcinomas reveal MCCs upon histological examinations. The high correlation between the appearance of the microcalcification clusters and the diseases shows that the CAD (computer aided diagnosis) systems for automated detection microcalcifications will be very useful and helpful for breast cancer control.In this paper, the contrast enhancement methods are studied for improving image qualities and detecting microcalcifications more easily for radiologist. Three methods, i.e, the top-hat method, the conventional partial wavelet reconstruction method, the weighted wavelet coefficient partial reconstruction method, are applied for image enhancement. The results of the three methods are respectively evaluated by the contrast improvement index. The result contrast improvement indexes demonstrate that the weighted wavelet coefficient partial reconstruction method is a more effective way to enhance microcalcifications than the others.Automated detection of microcalcifications is the main content of this dissertation. Our method consists of three steps. Firstly, microcalcifications are enhanced by the weighted wavelet coefficient partial reconstruction method. Secondly, potential microcalcifications are identified by multi-tolerance region growing algorithm. In order to get seeds for region growing, adaptive global grey-level thresholding is applied to the enhanced image. The threshold level is based on the grey-level histogram of the whole image and is chosen such that 98% of all pixels in the enhanced image are set to be background. If several microcalcifications are closeto each other, adaptive global-grey threshold may not be able to separate them correctly. This problem could be solved by combining morphological erosion and conditional thresholding. The point with the max grey level in every four-connected region of the binary image is selected as a seed. Thirdly, four properties including area, compactness, elongation and contrast of potential microcalcifications are calculated. Both backpropagation neural network and support vector machine are used as classifier to detect microcalcifications from potential ones. Finally the remaining signals in the processed image are then grouped or clustered by passage of a 3.2×3.2mm2 box over the image.