Objective:

To evaluate the diagnostic accuracy of a clinically interpretable AI model in distinguishing early-stage breast cancer from benign lesions on MRI.

Key Findings:

• CBM achieved an AUC of 0.92 ± 0.01 in the test set, comparable to a black-box model with AUC of 0.93 ± 0.01.
• In external validation, CBM maintained an AUC of 0.93, overall accuracy of 86%, and precision of 89%.
• CBM demonstrated diagnostic accuracy of 89% in distinguishing benign from malignant lesions, outperforming seven of eight radiologists.
• Radiologist accuracy improved from 71% to 79% without assistance to 77% to 91% with CBM assistance.
• Inter-reader agreement improved, and clinical decision-making was positively affected, with 22% of suspicious benign lesions correctly downgraded.

Interpretation:

The interpretable AI model can achieve high diagnostic performance while aligning with clinical reasoning and workflow, enhancing radiologist accuracy and efficiency.

Limitations:

• Retrospective study design and simulated reading setting may not reflect real-world workflows.
• Model depends on physician-provided lesion localization and wasn't evaluated for screening use.
• External validation was limited, and performance in broader clinical populations remains uncertain.
• Potential for misclassification of small or atypical lesions without characteristic imaging features.

Conclusion:

The CBM provides a versatile framework for classifying early breast cancer and benign lesions, with significant implications for improving diagnostic accuracy in clinical practice.