Evaluation and Modelling of Detection, Location and Discrimination of Lesions in Mammograms
Previously we described mass detection results for human and model observers using mammogram and filtered noise backgrounds with the same power spectrum. The results were unusual in that threshold amplitudes (contrasts) increased as mass size increased. We were able to predict these results using mathematical observer models. When observers had to search for the mass, human thresholds were identical for both backgrounds. This suggests that breast structure in the image is equivalent to random noise. In mass discrimination experiments, thresholds for masses larger than 4 mm were about 1.5 to 2 times higher than for detection. We conclude that detection, localization and discrimination of masses in mammograms are limited by statistical variation in normal breast structure in the image, rather than imaging system noise.
KeywordsImage Noise Observer Model Spatial Frequency Channel Breast Structure Specimen Radiograph
Unable to display preview. Download preview PDF.
- 1.Eckstein, M., Abbey, C., and Bouchud, F.: A practical guide to model observers for visual detection in synthetic and natural noisy images. In: Handbook of Medical Imaging. Physics and Psychophysics, Beutel, J., Kundel, H., van Metter, R.: (eds.) Vol. 1. SPIE Press, Bellingham WA, (2000) 593–628.CrossRefGoogle Scholar