AI False Negatives in Mammography Slash Radiologists' Sensitivity from 71% to 39%

When an AI tool incorrectly suggested a mammogram was normal, radiologists' median sensitivity fell from 71% to 39%, per a Radiology study. Eye-tracking showed less scrutiny of those exams, highlighting automation bias.

• For false positive AI suggestions, specificity was higher (39% vs 21% unassisted).
• True- and false-positive AI prompts increased median read time from 25 seconds to 34 seconds.
• Shorter fixation durations on lesions deemed normal by AI indicate less visual scrutiny.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.