AI Finds 40% of Interval Cancers on Prior False-Negative Mammograms, Mostly Visible Lesions

An FDA-approved AI tool detected 40% of interval breast cancers on false-negative mammograms: 81% of visible lesions vs 7% of occult ones (p<0.005). Detection correlated with visibility (OR 53.1), symptoms, and higher T stage.

• Interreader agreement for retrospective visibility classification was substantial (κ=0.69).
• AI missed no minimal actionable lesions (100% detected) and most minimal nonactionable ones (61%), but detected none of the 14 true interval cancers.

Automated summary

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