Artificial Intelligence Applied to Breast MRI for Improved Diagnosis

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate differentiation of malignant from benign breast lesions on MRI is crucial for effective patient treatment.
• Artificial intelligence (AI) offers potential for enhancing the interpretation of complex imaging features.

Purpose of the Study:

• To assess if an AI system improves radiologists' diagnostic performance in distinguishing cancer from non-cancer on dynamic contrast-enhanced (DCE) breast MRI.
• Comparison of AI-assisted interpretation versus conventional software for breast MRI analysis.

Main Methods:

• Retrospective reader study involving 19 breast imaging radiologists interpreting 111 DCE breast MRI examinations.
• Readers performed interpretations twice: once with conventional software and kinetic maps, and again with added AI analytics.
• Diagnostic performance was evaluated using receiver operating characteristic (ROC) analysis, with the area under the curve (AUC) as the primary metric.

Main Results:

• The average AUC across all readers improved from 0.71 to 0.76 (P = .04) when using the AI system.
• Sensitivity improved for BI-RADS category 3 lesions (90% to 94%) but not for category 4a.
• Specificity showed no significant difference for either BI-RADS category 4a or category 3.

Conclusions:

• The integration of an AI system enhances radiologists' diagnostic performance in differentiating benign and malignant breast lesions on MRI.
• AI-powered tools show promise in improving the accuracy of breast cancer diagnosis through MRI.