Framing the fallibility of Computer-Aided Detection aids cancer detection

Area of Science:

• Medical Imaging
• Human-Computer Interaction
• Cognitive Psychology

Background:

• Computer-Aided Detection (CAD) systems aim to enhance cancer detection in mammograms.
• Previous research indicates accurate CAD improves detection, while inaccurate CAD increases missed cancers and false alarms due to over-reliance.

Purpose of the Study:

• To investigate if framing statements about CAD fallibility can mitigate the over-reliance effect.
• To determine if this approach preserves the benefits of CAD in cancer detection.

Main Methods:

• Three experiments were conducted to test the impact of framing CAD fallibility.
• Experiment 1 framed CAD benefits or costs; Experiment 2 used stronger warnings about CAD costs.
• Experiment 3 involved a lower cancer prevalence target.

Main Results:

• Framing had no effect in Experiment 1.
• A stronger warning in Experiment 2 significantly reduced the over-reliance effect.
• Similar mitigation of over-reliance was observed in Experiment 3 with lower prevalence.

Conclusions:

• Over-reliance on CAD technology is a significant concern in medical imaging.
• Framing and instructional sets highlighting CAD fallibility can effectively mitigate over-reliance.
• This approach helps maintain CAD's benefits while improving diagnostic accuracy.