Artificial Intelligence Flags Interval Breast Cancers Visible at Screening

FRIDAY, May 23, 2025 (HealthDay News) — Artificial intelligence (AI) more often flags and accurately localizes interval breast cancers (IBCs) that are visible at screening than true interval or occult cancers, according to a study published online April 18 in the Journal of the National Cancer Institute.

Tiffany T. Yu, M.D., from the University of California, Los Angeles, and colleagues identified IBCs diagnosed <12 months after a negative mammogram from digital mammography and digital breast tomosynthesis screening mammograms acquired from 2010 to 2019 at a U.S. tertiary care academic center. IBCs were classified retrospectively as missed–reading error, minimal signs–actionable, minimal signs–nonactionable, true interval, occult, or missed–technical error. Risk scores of 1 to 10 were assigned to the negative index screening mammograms by a deep-learning AI tool, with scores ≥8 considered flagged.

A total of 148 IBCs were identified in 148 women from 184,935 screening mammograms. Of these, 26, 24, 22, 17, 6, and 5 percent were minimal signs–actionable, occult, minimal signs–nonactionable, missed–reading error, true interval, and missed–technical error, respectively. The researchers found that AI scored 131 mammograms (17 errors excluded). Exams with missed–reading errors, minimal signs–actionable, and minimal signs–nonactionable were most frequently flagged by AI (90, 89, and 72 percent, respectively). Mammographically visible types were more accurately localized by AI than nonvisible types (35 to 68 percent versus 0 to 50 percent, respectively).

“AI may facilitate detection of the mammographically visible IBC types (missed–reading error and minimal signs) at time of screening and possibly reduce the IBC rate to comprise primarily true interval cancers,” the authors write.

The study was partially funded by EarlyDiagnostics via a grant provided to one author.

Abstract/Full Text (subscription or payment may be required)