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

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Screening digital breast tomosynthesis (DBT) lacks long-term outcome data.
• Interval cancer rate serves as a surrogate for patient outcomes in DBT screening.
• Evaluating AI performance is crucial for improving screening efficacy.

Purpose of the Study:

• To assess an artificial intelligence (AI) algorithm's ability to detect and localize interval cancers in screening DBT.
• To validate the AI algorithm's diagnostic threshold using various screening DBT examination outcomes.
• To determine the potential of AI in reducing interval cancer rates.

Main Methods:

• Retrospective analysis of screening DBT examinations preceding interval cancer diagnoses (February 2011-June 2023).
• Application of a U.S. Food and Drug Administration-cleared AI algorithm to mark lesions and assign scores (0-100).
• Independent radiologist review of AI-positive examinations (score ≥10) and comparison of imaging/clinicopathologic features.

Main Results:

• AI correctly localized 32.6% (73/224) of interval cancers in retrospective screening DBT evaluations.
• Interval cancers detected by AI were associated with larger size and axillary lymph node positivity.
• AI demonstrated high accuracy in identifying true-positive (84.4%), true-negative (85.9%), and false-positive (73.3%) cases at a threshold score of ≥10.

Conclusions:

• AI accurately localized a significant proportion of interval cancers in retrospective screening DBT analysis.
• The AI algorithm shows potential for decreasing the interval cancer rate in breast cancer screening.
• Further validation and integration of AI may enhance screening DBT performance and patient outcomes.