An Empirical Investigation Of Variability In Contrast-Detail Diagram Measurements

The contrast-detail (C-D) diagram has been used widely yin the evaluation of imaging systems including screen-film combinations, computed tomography, and digital radiography. Despite its apparent simplicity and efficiency in generating observer performance data, the relationship between this method and the more rigorous ROC or forced-choice procedures is not well understood. With a high-quality image processing/simulation system, we made images by superimposing digitally simulated square patterns on radiographic noise. We then studied the statistical variability in the C-D diagram due to within- and between-reader differences, and the effect of different numbers of image samples. Results indicate that the percent standard deviations due to each of these variations are about 10-20%. We also compared the visibility thresholds measured by the C-D diagram method and by an 18-alternative forced-choice (18AFC) procedure. Although the thresholds defined in these two methods are fundamentally different, we found that the threshold contrasts measured by the two procedures have similar values.