Breast cancer is clearly a major public health concern. Currently, the most effective method of detecting breast cancer is mammography. However, while most mammographic abnormalities are benign, they must all be confirmed by additional studies which may include additional mammograms, ultrasound, and biopsy. A computer aided diagnostic (CAD) aide, which can decrease the uncertainty inherent in the evaluation of mammograms, offers the potential of sparing many women the trauma involved in undergoing biopsies to confirm a diagnosis. This SBIR proposal, submitted by Technology/Engineering Management, Inc., describes a research program to establish the feasibility of applying a linear programming (LP) based pattern classifier to the computer aided diagnosis of breast cancer. The ultimate goal of this research is to lay the foundation for a software-based advisor that would serve as a decision aide to the human diagnostician in assessing electronic images of breast tumors, thereby enabling a totally noninvasive diagnosis. Working in conjunction with a research team from the University of Chicago Department of Radiology (UCDR), we plan to evaluate the efficacy of the LP approach to discriminating malignant from benign lesions and compare it to a mammography diagnostic aide that relies on an artificial neural network (ANN) based classifier.
This research will form the foundation for a computer aided diagnostic (CAD) system that can be used by radiologists to assist in the diagnosis of mammographic abnormalities. The technique can also be applied to other imaging procedures including magnetic resonance imaging and positron emission tomography.
