This proposal describes the development of an artificial neural network (ANN) for breast cancer diagnosis. The system is trained to aid in the decision to biopsy those patients that have suspicious mammographic findings. This system will decrease the variability and increase the specificity of the decision to biopsy. the decision to biopsy is a two stage process: 1) the mammagrapher views the mammagram and determines the presence of absence of image features such as caclcifications and masses, 2) these features are merged to form a diagnosis. A recent study found that 52% of missed breast cancers are due to errors at the decision step. About 80% of the biopsies that are performed are benign. AxIS will provide a service to significantly improve this performance. The clinician reads a mammagram, transmits the findings to AxIS where the trained ANN returns a prediction of benign of malignant in less than one second. This prediction will be transmitted to the clinician who can then include this information in the medical decision.
The service provided by AxIS will dramatically reduce the error rate for the decision to biopsy. Preliminary results indicate that use of a computer aid could decrease by half of the number of benign cases that are biopsied without decreasing the number of malignancies found for a savings of $335million per year. Clients include practicing clinicians and managed care organizations.
