Figure 1.
Error rates for rare targets (red bars, ∼1% prevalence) and common targets (green bars, 50% prevalence) for two types of errors, false negatives and false positives.
The dark colored bars represent data average over all 14 observers. The light red bars represent low prevalence average errors (false negatives and false positives) for the six observers who participated in both arms of the study (low and high prevalence). The light green bars represent high prevalence average errors (false negatives and false positives) restricted to the cases that the six high prevalence observers did not also see during the low prevalence arm of the study. Regardless of these filtering of the data, low prevalence, false negative errors are markedly higher than high prevalence false negative errors.
Figure 2.
One of the seven cases not seen at low prevalence but seen by all 6 readers at high prevalence.
This is a case of a 56-year-old woman whose test screening mammogram was presented with a prior mammogram taken two years earlier. The case was rated level 5 of difficulty, the cancer was detected on the original screening and the lesion type is calcifications measuring 10 mm in size with the pathology of DCIS with microinvasion. The parenchymal density is less dense. a) MLO (top 2 images) and CC (bottom two images). b) Lesion in right upper quadrant. Magnification view shows pleomorphic calcifications in a segmental distribution.