AI to Reduce the Interval Cancer Rate of Screening Digital Breast Tomosynthesis

• Published in: Radiology Vol. 316; no. 1; p. e241050
• Main Authors: Bahl, Manisha, Langarica, Saul, Lamb, Leslie R, Kniss, Ariel S, Do, Synho
• Format: Journal Article
• Language: English
• Published: United States 01.07.2025
• Subjects: Adult; Aged; Algorithms; Artificial Intelligence; Breast - diagnostic imaging; Breast Neoplasms - diagnostic imaging; Early Detection of Cancer - methods; Female; Humans; Mammography - methods; Middle Aged; Radiographic Image Interpretation, Computer-Assisted - methods; Retrospective Studies; Sensitivity and Specificity
• ISSN: 1527-1315
• DOI: 10.1148/radiol.241050

Background Given the lack of long-term outcome data for screening digital breast tomosynthesis (DBT), the interval cancer rate is a commonly used surrogate for patient outcomes. Purpose To evaluate the performance of an artificial intelligence (AI) algorithm in detecting and localizing interval cancers at screening DBT and to validate the diagnostic threshold of the algorithm by analyzing its performance across interval, true-positive, true-negative, and false-positive screening DBT examinations. Materials and Methods Screening DBT examinations performed immediately before confirmed interval cancer diagnoses between February 2011 and June 2023 at an academic institution were retrospectively analyzed by a U.S. Food and Drug Administration-cleared AI algorithm. Lesions marked by AI on DBT sections were assigned a score from 0 to 100, with the examination-level score reflecting the highest lesion score. AI-positive examinations (score ≥10) were independently reviewed by two breast imaging radiologists to determine whether AI annotations corresponded to the site of the subsequent cancer. Imaging and clinicopathologic features were compared between interval cancers detected and not detected by AI using the Wilcoxon signed rank test and Fisher exact test. Using the same threshold score, 1000 true-positive, true-negative, and false-positive screening DBT examinations were also analyzed by AI. Results Among 224 interval cancers in 224 women (mean age, 61 years ± 14 [SD]), AI correctly localized 32.6% (73 of 224) of the cancers at retrospective evaluation of the screening DBT examinations. Features associated with interval cancers detected by AI (versus not detected) included larger size at surgical pathology (37 vs 22 mm; < .001) and axillary lymph node positivity (41.3% [26 of 63] vs 22.8% [28 of 123]; = .01). Using the same threshold (score of 10 or greater considered positive), AI correctly localized 84.4% (282 of 334) of screening-detected (true-positive) cancers and correctly categorized 85.9% (286 of 333) and 73.3% (244 of 333) of true-negative and false-positive cases, respectively, as negative. Conclusion AI correctly localized nearly one-third of interval cancers at retrospective evaluation of screening DBT examinations and has the potential to decrease the interval cancer rate. © RSNA, 2025 See also the editorial by Lee and Milch in this issue.