ARTIFICIAL INTELLIGENCE applied to breast MRI has demonstrated significant improvements in diagnostic accuracy and specificity, offering a potential solution to long-standing challenges in the assessment of indeterminate breast lesions.
Diagnostic Challenges in Breast MRI
Breast MRI is widely used to support breast cancer diagnosis, yet its clinical value is limited by high false-positive rates and substantial inter-reader variability. These challenges are particularly pronounced for lesions categorised as Breast Imaging Reporting and Data System category four, which frequently lead to unnecessary biopsies. Variability in radiologist interpretation further complicates decision-making and can affect patient management.
To address these limitations, researchers developed the BI-RADS four Lesions Analysis System, an artificial intelligence platform designed to analyse the rich spatiotemporal data generated by dynamic contrast-enhanced MRI. The system leverages foundation models to improve lesion characterisation and support more consistent clinical interpretation.
Performance And Clinical Accuracy
The artificial intelligence system was developed and evaluated using a multicentre dataset comprising 2,803 lesions from 2,686 female patients. Across analyses, the system demonstrated strong diagnostic performance, achieving areas under the curve ranging from 0.892–0.930. Importantly, the system significantly outperformed radiologists in specificity, achieving 0.889 compared with 0.491, highlighting its ability to reduce false-positive findings in breast MRI.
When integrated into clinical interpretation, the system significantly improved diagnostic accuracy for both senior and junior radiologists. Assisted readings were associated with a 27.3% reduction in false-positive rates, suggesting fewer patients would be referred for unnecessary biopsy. The use of artificial intelligence also reduced inter-reader variability by 24.5%, addressing a key source of inconsistency in breast MRI reporting.
Implications For Precision Breast Cancer Care
Beyond binary classification, the system enables refined risk stratification by assigning lesions to BI-RADS four subcategories A, B, and C. This granular assessment supports more personalised risk evaluation and may help guide more appropriate clinical management pathways.
Overall, these findings indicate that artificial intelligence-assisted breast MRI interpretation could play a meaningful role in precision breast cancer management. By improving specificity, reducing variability, and enhancing radiologist performance, this approach offers a practical tool to optimise diagnostic workflows and improve patient outcomes while minimising unnecessary interventions.
Reference
Liang Y et al. An interpretable AI system reduces false-positive MRI diagnoses by stratifying high-risk breast lesions. Nature Communications. 2026; https://doi.org/10.1038/s41467-026-69212-7.
