| Breast cancer is one of the most common malignant diseases among women, the incidence of which is increasing with a high rate. There is clear evidence showing that it is important to give patients early diagnose and treatment, which can survive more of them. Mammography has become the most effective way to detection of breast cancer, and it is sensitive to clustered micro-calcification which is the key characteristic of early breast tumors. Since the information in the mammogram can not be identified entirely, some symptom may be missed by an expert. We intend to use the computer aided diagnose to help experts before they read mammograms, by listing the areas likely to include micro-calcification.Two steps in detection of micro-calcification are as follows: 1) selection of the region of interest (ROI) from the whole breast area, which is the doubtful micro-calcification region; 2) detection of micro-calcification in all ROIs. This thesis gives a method to identify the breast area from the whole mammogram using C-means clustering algorithm, which has a perfect effect and decreases amounts of computation in the following detection. Then each breast region must be decomposed using Daubechies wavelets, and be considered as an ROI if it suffices the two-threshold selection, when the two thresholds are dynamically created in this thesis. To reduce the false positive (FP), ROIs are filtered with difference of Gaussian (DOG) to detect micro-calcification, and every pixel will be identified through a threshold to make sure whether it is a micro-calcification or not. Finally, the position of micro-calcification pixel is displayed. After that, the FP goes down, while holding a high sensitivity.In this thesis, 24 DICOM mammograms have been processed, holding the sensitivity of 83.39% and the FP of 2.30% in ROI selection, the sensitivity of 83.39% and 3.58 FP regions per image in detection of micro-calcification in ROIs, which reach a high level and are recognized by experts in breast cancer. |
PDF Full Text Request