Abstract
Breast cancer detection has undergone a massive technological shift—moving from mammography, a process that takes a two-dimensional (2D) image of breast tissue, to tomosynthesis, a technique that creates a segmented-three-dimensional (3D) image. When observers search with tomosynthesis/ 3D displays, they are more accurate and make fewer false alarms compared to mammography/ 2D displays (e.g., Adamo et al., 2018; Ciatto et al., 2013). Because the incidence of breast cancer is low (127 out of 100,000 women in the U.S.), it is critical to determine if 3D search benefits persist with low-prevalence searches compared to high-prevalence searches given that low-prevalence searches typically yield worse search performance (Wolfe et al., 2005). 108 undergraduate students from the University of Central Florida were asked to search in both 3D and 2D environments previously used by Adamo et al., (2018). Importantly, the search program used emulated tomosynthesis but did so with simplified stimuli that were accessible to non-radiologist observers. In 3D, observers traversed throughout a sphere, moving from one search display “slice” to the next and searching for a target “T” amongst distractor “Ls.” In half of the trials, the 3D sphere was “compressed” into a 2D display akin to how a breast image can be viewed using tomosynthesis or as a mammogram. Participants were divided between 3 conditions—10%, 50%, or 100% target prevalence. A series of 2x2 ANOVAs revealed a main effect of condition (2D vs. 3D)—participants had significantly more hits, fewer false alarms, and better sensitivity when searching in 3D compared to 2D. Furthermore, as prevalence increased, performance improved—better hit rate, fewer false alarms, and improved sensitivity. The results suggest that the benefits of searching in 3D environments persist at lower prevalence, but do not counteract the reduction in search performance traditionally observed in low-prevalence searches.