Assessing task performance in FFDM, DBT, and synthetic mammography using uniform and anthropomorphic physical phantoms

Purpose: The purpose of this study is to quantify the differences in detectability between full field digital mammography (FFDM), digital breast tomosynthesis (DBT), and synthetic mammography (SM) for challenging, low contrast signals, in the context of both a uniform and an anthropomorphic, textured phantom. Methods: Images of the phantoms were acquired using a Hologic Selenia Dimensions system. Images were taken at 50%, 100%, and 200% of the dose delivered under automatic exposure control (AEC). Low-contrast disks, created using an inkjet printer with iodine-doped ink, were inserted into the phantom. The disks varied in diameter from 210 to 630 mu m, and in local contrast from 1.1% to 2.8% in regular increments. Human observers located the disks in a 4 alternative forced choice experiment. Proportion correct (PC) was computed as the number of correct localizations out of the total number of tries. Results: Overall, scores from FFDM and DBT were consistently greater than scores from SM. At an exposure corresponding to the AEC setting, mean PC scores for the largest disks with the uniform phantom were 0.80 for FFDM, 0.83 for DBT, and 0.66 for SM, with the same rank ordering at other doses. Scores were similar but lower for the nonuniform background. At an exposure twice the AEC setting, however, the difference between uniform and nonuniform scores was most pronounced for DBT alone. Differences between scores for FFDM and SM were statistically significant, while those between FFDM and DBT were not. Scores were used to compute the minimum contrast level needed to reach 62.5% detection rate. The minimum contrast for SM was 36%-81% higher compared to FFDM or DBT, in either background. Conclusions: This study shows that an anthropomorphic phantom and lesions inserts may be used to conduct a reader study. Detectability was significantly lower for synthetic mammography than for FFDM or DBT, for all conditions. Additionally, observer performance was consistently lower for the anthropomorphic phantom, indicating the greater challenge due to anatomical background. Because of this, it may be important to use realistic phantoms in observer studies in order to draw conclusions that are more clinically relevant. (C) 2016 American Association of Physicists in Medicine.