Katharine E Sabo¹ (), Dawn M Sarno¹; ¹Clemson University

Computer-aided detection (CAD) systems are essential tools for aiding radiologists in assessing breast images for cancer. Despite their widespread use, limited research has compared the various types of CAD systems, particularly for digital breast tomosynthesis (DBT) imaging, which is anticipated to surpass mammography in use. Additionally, calls for including global classifications in CAD systems, particularly the Breast Imaging Reporting and Database System (BI-RADS), suggest such ratings may enhance diagnostic accuracy (Qian et al., 2015). This study examined three existing CAD system types (binary, analog, and interactive); a novel CAD system with a global classification (BI-RADS); and a control with no CAD system. Undergraduate students (N = 50) were randomly assigned to these conditions and tasked with identifying a target (T) among distractors (Ls) in mock DBT displays. For each trial, stimuli were randomly distributed across 13 layers to mimic the pseudo-3D images of DBT sets. Assistance provided to participants in the automation conditions varied by CAD type: binary CAD circled likely targets; BI-RADS CAD supplied a rating and descriptor indicating the presence of a target; analog CAD circled potential targets and provided probability estimates; and interactive CAD (iCAD) required participant activation on each trial to reveal an analog CAD overlay. Results revealed that participants in the binary and analog conditions exhibited an average 39% increase in hit rate and a 16.59s decrease in response times compared to the control, BI-RADS, and iCAD conditions. Notably, participants in the BI-RADS and iCAD conditions did not significantly differ from the control condition or each other. While iCAD and BI-RADS are considered advancements in CAD technology, this research suggests that further investigation into their influence on search performance is necessary before replacing the widely used binary and analog CAD systems.