A BAYESIAN NETWORK TO ASSIST MAMMOGRAPHY INTERPRETATION.

Mammography is a vital screening test for breast cancer because early diagnosis is the most effective means of decreasing the death rate from this disease. However, interpreting the mammographic images and rendering the correct diagnosis is challenging. The diagnostic accuracy of mammography varies with the expertise of the radiologist interpreting the images, resulting in significant variability in screening performance. Radiologists interpreting mammograms must manage uncertainties arising from a multitude of findings. We believe that much of the variability in mammography diagnostic performance arises from heuristic errors that radiologists make in managing these uncertainties. We developed a Bayesian network that models the probabilistic relationships between breast diseases, mammographic findings and patient risk factors. We have performed some preliminary evaluations in test cases from a mammography atlas and in a prospective series of patients who had biopsy confirmation of the diagnosis. The model appears useful for clarifying the decision about whether to biopsy abnormalities seen on mammography, and also can help the radiologist correlate histopathologic findings with the mammographic abnormalities observed. Our preliminary experience suggests that this model may help reduce variability and improve overall interpretive performance in mammography. [ABSTRACT FROM AUTHOR]