Is there a safety-net effect with computer-aided detection?

Area of Science:

• Medical Imaging
• Radiology
• Human Factors in Medicine

Background:

• Computer-aided detection (CAD) systems assist in interpreting screening mammograms by highlighting suspicious regions.
• A potential 'safety-net' effect exists where the preliminary unaided search may be less thorough due to the subsequent CAD review.

Purpose of the Study:

• To investigate the 'safety-net' effect in computer-aided detection (CAD) during mammogram interpretation.
• To assess how the availability of CAD prompts influences the initial visual search for microcalcification clusters.

Main Methods:

• A visual search experiment was conducted with 52 nonexpert observers simulating breast screening with CAD.
• Participants searched 100 synthetic images for microcalcification clusters under both prompted (CAD) and unprompted (no-CAD) conditions.
• Eye-tracking data and behavioral sensitivity were collected to analyze image coverage and detection accuracy.

Main Results:

• Image coverage was significantly lower in the initial search phase when CAD prompts were available compared to the no-CAD condition.
• Observer sensitivity was significantly higher for targets identified by CAD compared to the same targets in the no-CAD condition.
• No significant difference in sensitivity was observed for targets *not* marked by CAD between the CAD and no-CAD conditions.

Conclusions:

• The availability of CAD may influence the thoroughness of the initial unaided search, potentially reducing overall image coverage.
• CAD systems enhance sensitivity for their detected targets but may not improve detection of unprompted targets.
• Future CAD efficacy studies should consider and account for this potential 'safety-net' effect on initial search behavior.