Likelihood Systems Can Improve Hit Rates in Low-Prevalence Visual Search Over Binary Systems

Area of Science:

• Human-Computer Interaction
• Cognitive Psychology
• Information Science

Background:

• Low target prevalence in visual search tasks, such as medical imaging and security screening, often results in low hit rates.
• Traditional binary decision support systems generate numerous false alarms in low-prevalence settings, limiting their effectiveness.
• Likelihood-based systems offer enhanced certainty information, potentially improving performance without exacerbating false alarms.

Purpose of the Study:

• To compare the performance of binary versus likelihood decision support systems in low-prevalence visual search tasks.
• To determine if likelihood systems can increase hit rates without a corresponding rise in false alarms.

Main Methods:

• A simulated medical search task was employed with a low target prevalence.
• Participants completed the task across two sessions, each supported by either a binary or a likelihood decision support system.
• Interaction involved sequentially uncovering stimulus parts using a mouse interface.

Main Results:

• Participants achieved higher hit rates when using likelihood decision support systems compared to binary systems.
• The observed increase in hit rates with likelihood systems occurred without a significant increase in false alarm rates.
• This suggests likelihood systems are effective in improving detection accuracy.

Conclusions:

• Likelihood decision support systems present a viable solution for addressing the challenges of low-prevalence visual search.
• These systems can enhance transparency by providing clear indications of system certainty.
• The findings suggest that providing simple, interpretable certainty information is key to improving performance in domains like radiology.