Channelized Hotelling observer assessing microcalcification detectability on 2D mammography: A first application to study the impact of tube voltage

This work was aimed to assess the detectability of microcalcifications as a function of tube voltage in 2D mammography using an in-house developed Channelized Hotelling Observer (CHO). The images of a 3D structured phantom, containing acrylic beads of different diameters in water and inserted signals (microcalcifications, spiculated and non-spiculated mass models) were assessed. Images had been acquired on a Siemens Mammomat Inspiration system at 5 different tube voltage levels (28-32 kVp) under AEC setting. The detectability of microcalcifications was analyzed in terms of percentage correctly detected signals (PC) as well as diameter threshold (d_tr), using the CHO and human observers. Two Laguerre-Gauss and 8 Gabor channels were included in the two-layers CHO. The model and human observer results were retrieved from a 4-alternative forced choice (4-AFC) study. At 28 kVp and 32 kVp, the diameter thresholds d_tr of microcalcifications together with their 95% confidence intervals (CI95) were 0.114 [0.110-0.118] mm and 0.119 [0.116-0.122] mm for the CHO, and 0.110 [0.093-0.127] mm and 0.123 [0.111-0.134] mm for the human observers. The Pearson correlation (r) between the PC values of model and human observer was more than 0.934. The in-house developed CHO and the human observer scores correlated very well for the application on 2D digital mammography acquired with different tube voltages. The overlapping range of CI95 of the d_tr for the tested kVp shows that the tube voltage setting does not significantly affect the detectability of microcalcifications neither by the CHO nor by the human observers.