MANAGEMENT OF UNCERTAINTY IN BREAST-CANCER GRADING WITH BAYESIAN BELIEF NETWORKS

OBJECTIVE: To examine the potential of different constructs of Bayesia n belief networks (BBN) to manage uncertainty in breast cancer grading . STUDY DESIGN: We developed four networks, two based on bloom-Richard son's and two on Helpap's grading systems. The function of the network s was based either on an expert's experience or frequency counts deriv ed from subjective grading of a large number of samples. The four BBNs were tested on 20 specimens, and the resulting final beliefs were com pared with the subjective gradings. RESULTS: The BBNs showed agreement with the subjective gradings in 60-85% of cases. Different constructs of BBNs, however, differed in their performance. The mean beliefs in frequency-based networks were slightly higher than in experience-based networks. In addition, as compared with the Bloom-Richardson-based ne tworks, the Helpap-based BBNs resulted in higher maximum beliefs but p roduced a larger fraction of discrepancies with the subjectively grade d cases. Depending on the type of network, 65-90% of the BBN grades we re associated with high beliefs. CONCLUSION: The results suggest that for reliable results, grading systems with more than three or four var iables may be necessary. When based on relevant information, BBNs seem to have the potential to become a valuable method of assisting the pa thologist in breast cancer grading.