Reliability of radiologists' first impression when interpreting a screening mammogram.

Area of Science:

• Radiology and Medical Imaging
• Human Perception and Cognition

Background:

• Radiologists can identify abnormalities in mammograms quickly via global processing.
• Initial impressions, or 'gist signals,' are formed rapidly from screening mammograms.

Purpose of the Study:

• To assess the reliability of radiologists' initial 'gist signals' in mammogram interpretation.
• To determine intra- and inter-observer reliability for these rapid impressions.
• To identify if a subset of radiologists demonstrates superior reliability and accuracy.

Main Methods:

• Thirty-nine radiologists provided initial impressions on mammograms twice, with each image viewed for 0.5 seconds.
• Intra-class correlation (ICC) and weighted Cohen's Kappa were used to measure reliability.
• Statistical tests (Mann-Whitney U) compared reliability metrics between 'Gist Experts' and other radiologists.

Main Results:

• Intra-reader reliability was poor to moderate (median ICC not specified, median Kappa 0.478).
• Only a minority of radiologists met minimum reliability standards (ICC ≥ 0.6).
• Inter-reader reliability was poor (ICC 0.31, Fleiss Kappa 0.106), indicating slight agreement.

Conclusions:

• Radiologists' initial 'gist signals' from rapid mammogram viewing are not reliably consistent.
• The absence of an abnormal gist does not reliably indicate a normal mammogram.
• Emphasizes the need for thorough 'discovery scanning' before concluding a visual search.