External validation of deep learning BI-RADS model for breast MRI: AUC 0.767 versus radiologist AUC 0.913

Deep learning BI-RADS model for breast MRI yielded an AUC of 0.767 (95% CI 0.693–0.830) vs radiologist AUC 0.913 (0.857–0.952) on external validation of 158 exams, a significant difference of 0.146 (P=0.0001). At a BI-RADS >3 threshold, it flagged 54.7% of benign exams as low-ri…

• Retrospective development on 12,421 breast MRI exams from three institutions; external validation on 158 exams from an independent outpatient center with different hardware and protocols.
• At a BI-RADS >3 threshold, the model identified 54.7% of benign examinations as low-risk, yielding a negative predictive value of 97.4%.
• Prospective evaluation is needed to determine whether this can safely improve workflow.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.