Comparisons of imaging systems through observer performance studies involve the selection of images, including those with subtle abnormalities. The selection of subtle abnormalities is typically performed in a subjective manner and is not quantitative. To assess the validity of such a selection process, we analyzed 163 ROC data sets. The difference in observer performance summary indices (Az) between ''typical'' and ''subtle'' cases were evaluated for each set. Average difference was found to be 0.10 (Az ''typical'' - Az ''subtle''), and in only 4 instances of 163 comparisons, observer performance for the subtle cases increased by 0.01 ((Delta) Az). We conclude that although it is not quantitative, this approach to the selection of subtle abnormalities is appropriate for imaging systems comparisons using ROC- type studies.
|