Evaluation of a deep learning-based software to automatically detect and quantify breast arterial calcifications on...

Area of Science:

• Radiology
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Breast arterial calcifications (BAC) are associated with cardiovascular disease.
• Current methods for BAC assessment are manual and time-consuming.
• AI offers potential for automated and efficient analysis of medical images.

Purpose of the Study:

• To evaluate an artificial intelligence (AI) software for automatic detection and quantification of BAC.
• To compare AI-based BAC scoring with manual radiologist scoring.
• To assess the diagnostic performance of AI-detected BAC for predicting coronary artery calcification (CAC).

Main Methods:

• Retrospective study of 502 women with mammography and CT scans (2009-2018).
• Deep learning software generated a BAC AI score (0-10 points).
• Comparison with manual BAC scores and manual CAC scores (12-point scale).

Main Results:

• BAC AI score strongly correlated with manual BAC score (r=0.83).
• AI-based BAC score (≥5) showed 32.7% sensitivity and 96.1% specificity for detecting marked CAC (≥4).
• Area under the ROC curve (AUC) for marked CAC detection was 0.64.

Conclusions:

• Automated BAC AI scoring strongly correlates with manual scoring.
• AI-based BAC assessment can be a valuable tool for mammography reports.
• This AI tool may enhance awareness of cardiovascular risk in women.