Study On Detection Methods Of Calcification In Mammography

Early diagnosis of breast cancer is the key for reducing mortality, so many researcher have dedicated themselves to the study on calcification detection methods in mammography. Calcification detection, which is part of computer-assisted diagnosis (CAD), is a difficult and a key step in diagnosis early breast cancer. So, with calcification character, such as smallness in size, different size and shape, very small intensity difference between suspected lesion region and surrounding tissue, partly tissue considered easily as calcification mistakenly and so on , this dissertation attempt to develop a integrated detection methods with a few techniques including general image technique, wavelet transform and neural network etc.The detection process in this paper mainly includes image pre-processing, calcification segmentation, feature extraction, recognizing and eliminating false positive calcification, classifying between benign and malignant calcification. The difficulty in this process is full calcification segmentation in image. Tree-structure non-linear filter (TSF) has been used to eliminate noise of image in pre-processed stage. Calcifications is seen as a tiny grain in image, and has high gray value than surrounding pixels, and look like singular points in function or high frequency in signal. According to the calcification character, a synthesized method has been used to segment calcifications of mammography in segmentation stage. Firstly, based on difference-image technique and local threshold methods, calcifications have been segmented to produces a binary image of calcification. Afterwards, wavelet transform has been used to generating detail sub-image on mammography and the sub-image is used to gain a calcification binary image. Secondly, above two binary image is done by "and" operation to generate a final binary image. This stage can reduce false positive calcification. With this binary image, calcification position and seeds is confirmed, and later region growing method has been adopted to segment full calcification in mammography.In order to reduce more false positive calcifications, BP neural network is used to recognize and eliminate false positive calcifications. All features extracted in this paper includes shape feature, intensity feature, cluster description feature. Later shape features will be analyzed.Calcification detection algorithm in this paper is tested by a number of mammography. Initial results suggest that the proposed methods for calcification detection is feasible and can segment full calcification and reduce false positive calcification. Thus expectant purpose is obtained.