Abstract
We have used phantom images to develop a digital method for assessing mammogram quality because it is more objective than visual evaluation. Using this method we analyzed images gathered in a regional survey of mammogram quality.
When our digital method of image quality assessment is used in a quality control process it is necessary to set a predetermined path/fail level. To determine the appropriate level we compared our quantitative measurements with the results of visual evaluation, the standard method, of image quality using 42 images of American College of Radiology accredited phantoms. The images were digitized and the signal-to-noise ratios (SNR) of calcifications (C ; Cl for the largest specks and C3 for the largest third specks) and masses (M ; M1 for the largest mass and M 3 for the largest third mass) were measured. The visual evaluations were performed by 14 readers-9 with substantial experience (group A) and 5 less experience (group B) who scored the images according to the Guidelines of the Japanese Association of Breast Cancer Screening. For each image, the group B readers gave under score than the group A readers. However, both groups showed large variations in scoring. The scores for calcifications and masses by readers in group A were mostly accurate when the SNRs of the test objects were higher than specific values : 13 in C1, 7.0 in C3 and 6.0 in M1. On a relative performance scale those values correspond to 0.75, 0.80 and 0.70, respectively.
We conclude that visual scoring of mammogram quality, even by experienced readers, results in large discrepancies. On the other hand, the path/fail level of image quality can be determined without ambiguity by quantitative measurement and we tentatively set the level to 0.7 to 0.80 on the relative performance scale.
