The goal of the proposed research is to develop computer-aided diagnosis (CAD) schemes to assist radiologists in diagnosing breast cancer from mammograms. We believe that CAD will improve radiologists' ability to interpret mammograms, so that both the number of missed cancers and the number of women unnecessarily sent to biopsy can be reduced. This proposal is specifically for the development of CAD schemes for the detection and the classification (benign versus malignant) of microcalcifications from digital mammograms. Under existing NIH funding, other research projects in our laboratory are directed at developing computer schemes for the detection and classification of breast masses. The schemes for both masses and microcalcifications will be tested together in the final years of this proposal.
The specific aims are: (1) To establish a large database of normal and abnormal (benign and malignant) mammograms, consisting of 2400 digitized screen-film mammograms (600 cases). (2) To develop an improved computer scheme for the automated detection of clustered microcalcifications by developing nonlinear filters, based on morphological operators, and advanced feature-extraction techniques, based on the radiographic features of microcalcifications and on the physical characteristics of the image receptor. Artificial neural networks will also be used to eliminate false-positive clusters detected by the computer scheme. (3) To develop a computer scheme for the classification of clustered microcalcifications. Two different approaches will be used -- artificial neural networks and rule-based techniques. Computer techniques to extract radiographic features of microcalcifications will be developed and these features will be used as input to the two classifiers. 4) To evaluate the ability of CAD to improve radiologists' accuracy in detecting and classifying breast lesions using ROC analysis. In addition, a pilot study for the clinical evaluation of the CAD schemes will be conducted.
Showing the most recent 10 out of 19 publications