Comparison of two psychometric functions in analyzing 4-AFC detection results using a task-based structured phantom for digital mammography

Abstract

Introduction Detectability of spiculated and non-spiculated mass models and microcalcifications, in a structured background phantom of acrylic spheres and water, was studied for digital mammography via the four-alternative forced-choice (4-AFC) paradigm. Purpose To compare the threshold diameters for mass and microcalcification detection using two types of psychometric curves applied on the contrast-detail data. Material and methods Ten images of the structured phantom were acquired on 23 digital mammography systems under automatic exposure control. The 4-AFC study was performed with in-house developed software and 5 medical physicists as observers. Percentage correctly detected lesions (PC) was obtained for each target type. Two psychometric curves (Weibull and Logistic functions) were applied to determine the threshold diameters and number of detected targets (at PC = 62.5%). Results Both fitting procedures were successfully applied on all systems. After averaging over 23 systems, the threshold diameters for spiculated and non-spiculated masses and microcalcifications, with the Weibull psychometric curves, were 4.0 ± 0.3 mm (R2 = 0.61 ± 0.08), 4.6 ± 0.7 mm (R2 = 0.51 ± 0.11), and 0.114 ± 0.003 mm (R2 = 0.91 ± 0.03) respectively, and the values of 4.0 ± 0.4 mm (R2 = 0.61 ± 0.08), 4.7 ± 0.6 mm (R2 = 0.50 ± 0.11), and 0.114 ± 0.003 mm (R2 = 0.91 ± 0.03) were obtained with the Logistic function. The numbers of detected targets assessed from the two functions were not different (3–4 spiculated masses, 1–2 non-spiculated masses, and 4 groups of microcalcifications). If the PC values were not completely covering the lower and upper than 62.5%, there will be possibility different results occurred from these two functions. Conclusion Weibull and Logistic curve fits gave similar and reproducible threshold diameters for the structured phantom.