In mammographic softcopy reading, assessment of contrast resolution is mainly performed with phantoms, including detection tasks with a homogeneous image background. For tasks in visual perception a processing hierarchy is assumed, where detection tasks represent the base level. The results of investigations based on detection tasks might not allow predictions on the sensitivity for recognizing low-contrast patterns in a situation with complex images. We introduce the MCS method (Mammographic Contrast Sensitivity) for determining the contrast sensitivity function (CSF) in mammograms. Gabor patterns and digits are used as visual targets. The observers have to cope with an orientation discrimination task for the Gabor patterns and an identification task for the digits. The contrast thresholds are measured by a psychophysical staircase procedure at six spatial frequencies up to 16 cycles per degree. A study with eight observers was performed to show the applicability of the MCS method. The results of the observer study with several mammographic cases show that the approach is applicable independent of the chosen images. The results for Gabor pattern targets were different from those with digits, both in overall sensitivity and in the shape of the contrast sensitivity function. Sensitivity to pattern recognition is thus not reliably predicted from the Gabor CSF, and a more complex target like a digit or a character should be preferred. The measurement of a contrast sensitivity function does not take more than 4 minutes. The results can be used to appraise the effects of viewing conditions with an aim of drawing conclusions for mammographic softcopy reading.

Date Published: 12 March 2009
PDF: 11 pages
Proc. SPIE 7263, Medical Imaging 2009: Image Perception, Observer Performance, and Technology Assessment, 726318 (12 March 2009); doi: 10.1117/12.810221

Published in SPIE Proceedings Vol. 7263:
Medical Imaging 2009: Image Perception, Observer Performance, and Technology Assessment
Berkman Sahiner; David J. Manning, Editor(s)