Overview

• Researchers retrospectively reviewed 1,376 digital breast tomosynthesis screenings and identified 224 interval cancers for algorithm evaluation.
• Lunit INSIGHT DBT v1.1.0.0 accurately localized 32.6% of cancers that were previously missed on initial DBT exams.
• In a separate analysis of 1,000 patients, the algorithm achieved 84.4% true-positive localization, correctly classified 85.9% of true-negative exams and marked 73.2% of false-positive cases as negative.
• The study employed lesion-specific evaluation to credit the AI only when it precisely pinpointed the cancer site, ensuring a rigorous assessment of sensitivity.
• The results indicate that integrating AI into DBT workflows could lower interval cancer rates, with larger and more aggressive tumors most likely to be detected, pending broader clinical validation.