Sensitive to a T: testing two types of computer-aided detection and two user response types in a visual search task

Spatial computer-aided detection (CAD) systems help radiologists detect tumours by marking suspicious regions in mammography images. When CAD cues are correct, observers find more tumours than when searching without CAD. However, when CAD cues are absent or incorrect, observers miss more tumours than without CAD. Kunar et al. (2017) suggest this is due to relying too much on CAD cues (the overreliance hypothesis). In addition to an overreliance on the information that CAD cues provide, attentional capture by CAD cues may aid or impair the observer. We conducted three visual search experiments where participants searched for a T among Ls embedded in noise to simulate mammography images. The first two experiments tested non-spatial CAD cues (removing attentional capture), with participants making present/absent (Experiment 1) or localisation (Experiment 2) responses. CAD increased overall performance only for present/absent responses. Experiment 3 used spatial CAD cues to compare the impact of response type (present/absent vs localisation). Miss rates were higher in the Present/Absent than the Localisation condition, suggesting this type of response encourages a less complete search than localisation. As radiologists use localisation, we suggest future research on medical imaging and CAD do the same. Changes in miss rates (compared to No CAD) in Experiment 2 demonstrate that observers are influenced by the information provided by CAD systems. A comparison of Experiments 2 and 3 suggests that attentional capture contributes to increasing overall performance. Thus, both information content and attentional capture by CAD appear to be responsible for altering observer behaviour. However, (relative to No CAD trials) lower false alarm and higher miss rates on trials where the CAD marks a distractor and fails to mark the target, suggest that attentional capture, rather than overreliance on the CAD cues, is responsible for miss errors on trials with incorrect CAD cues.