A study presents an interpretable mammography-based model to better stratify malignancy risk within BI-RADS 4 breast lesions and compares model variants.
Key Takeaways
- Retrospective model used 441 BI-RADS 4 lesions and found 165 malignancies
- Full mammography model achieved an optimism-corrected AUC of 0.841 versus 0.770 for BI-RADS alone
- Final predictors included BI-RADS 4B/4C, architectural distortion, shape and texture mammographic features