Related Experiment Video
Updated: Aug 5, 2026
Integrating Augmented Reality Tools in Breast Cancer Related Lymphedema Prognostication and Diagnosis
Published on: February 6, 2020
Concordance Between Radiologist and AI-based Volumetric Breast Density Assessments: Clinical and Economic
Jacob R Devine1, Alexander D M Withrow1, Sabina Choudhry2,1
1University of South Dakota Sanford School of Medicine.
Background:
Radiologists classify breast density using the Breast Imaging Reporting and Data System (BI-RADS) as required by the 2024 Mammography Quality Standards Act (MQSA), which mandates disclosure of breast density in all mammography reports. FDA-approved AI software now provides objective, reproducible breast density assessments to improve workflow and guide supplemental screening. This study examines concordance between AI-based systems (Volumetric density measurement by Volpara/Lunit) and radiologist assessments and explores potential clinical and economic implications.
Materials And Methods:
A retrospective cohort study with IRB approval was conducted at Avera Health using 54,819 mammograms between February 2024 and June 2025. Radiologists assessed breast density with the option to override AI-generated assessments. Density classifications were compared and categorized as "agreement" or "substantial change." Effects on Tyrer-Cuzick (TC) lifetime risk scores, patient referrals, and prospective healthcare costs were analyzed.
Results:
Overall, AI and radiologist assessments had agreement rates of 92-97% for BI-RADS B and C and 85-90% for A and D. Most discrepancies involved BI-RADS B and C categories, leading to misclassification of breast density and potential repercussions for patient care. Overestimation resulted in unnecessary procedures, radiation, costs, and anxiety, while underestimation risked delayed detection of aggressive cancers.
Conclusion:
Current AI software demonstrates concordance with radiologist assessments, though notable variability persists at critical diagnostic thresholds. Utilizing standardized protocols, periodic AI model updates, and radiologist oversight are crucial to minimizing misclassification, enhancing workflow efficiency, optimizing patient outcomes, and fostering trust in AI-assisted breast imaging.
