December 2022
Volume 22, Issue 14
Open Access
Vision Sciences Society Annual Meeting Abstract  |   December 2022
Incorrect Computer Aided Detection (CAD) marks lead to early quitting: A potential mechanism for poor CAD performance in clinical practice
Author Affiliations
  • Trafton Drew
    University of Utah
  • Anna Carissa Delos Reyes
    University of Utah
  • Jeff Moher
    Connecticut College
Journal of Vision December 2022, Vol.22, 3697. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Trafton Drew, Anna Carissa Delos Reyes, Jeff Moher; Incorrect Computer Aided Detection (CAD) marks lead to early quitting: A potential mechanism for poor CAD performance in clinical practice. Journal of Vision 2022;22(14):3697.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Prior work has shown that irrelevant distractors with an abrupt onset result in early quitting in a visual search task (Moher, 2020). Thus, while salient distractors often increase reaction time when a target is present, the same distractors elicit early quitting when no target is present. This may have important implications for applied tasks like medical screening, particularly in the context of CAD. In cancer screening, radiologists face an extremely difficult task where small differences could cost lives. There is a great deal of enthusiasm to help improve radiologist performance through AI-assisted cueing like CAD. Unfortunately, there are number of instances, such as breast cancer-screening, where CAD does not yield improved performance when deployed in the field. Prior work suggests that disappointing performance may be driven by attentional capture by CAD marks. The current examination sought to determine if the previously identified early quitting effect replicates in a situation more similar to CAD in screening radiology. Here, unlike Moher (2020), CAD marks sometimes highlighted targets (Ts embedded in 1/f noise) and appeared contemporaneously with the search array. Observers were divided into CAD and no CAD groups. We found that RTs greatly decreased (~500ms faster) on trials where there was an incorrect CAD mark on a distractor item (Ls). In addition, when a target was present and the CAD mark was on a distractor, accuracy was decreased relative to performance on the same targets in the No CAD condition. These results demonstrate that the previously identified phenomenon of early quitting in response to irrelevant distractor cues generalizes to a paradigm designed to emulate important aspects of breast cancer screening. More generally, these results suggest that early quitting may be an under-appreciated factor in understanding why CAD cues are less effective than expected in medical practice.


This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.