In the United States, breast cancer is the second leading cause of death in women. One out of eight women will develop breast cancer in their lifetime . Studies have indicated that early detection and treatment improve the chances of survival for breast cancer patients [2,3]. At present, mammography is the only proven method that can detect minimal breast cancers. However, 10-30% of the breast cancers that are visible on mammograms in retrospective studies are not detected due to various technical or human factors [4,5]. Double reading can reduce the miss rate on radiographic reading . It has also been shown that computer-aided diagnosis (CAD), in which a computer alerts radiologists to suspicious locations on the images during mammographic reading, can improve the detection accuracy significantly [7,8]. CAD is thus a viable cost-effective alternative to double reading by radiologists.
We are developing a computerized image analysis system to assist radiologists in mammographic interpretation. At present, we focus on the detection and classification of two of the most important mammographic indicators of breast cancers: masses and clustered microcalcifications. A schematic diagram of our CAD system is shown in Fig. 1. The mammograms for a patient are digitized by a high-resolution film scanner. The digitized mammograms are then processed by our automated detection programs to identify the regions containing suspicious microcalcifications or masses. In each region of interest (ROI), the identified lesion is analyzed by the appropriate classifier to estimate its likelihood of malignancy. The digitized mammograms will be displayed on the CAD workstation, as shown in Fig. 2. The locations of the detected lesions and their likelihood of malignancy will be superimposed on the displayed mammograms. The radiologist will read the original film mammograms and may use the computer information as a second opinion to make a diagnostic decision.