Including AI in diffusion-weighted breast MRI has potential to increase reader confidence and reduce workload.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Breast diffusion-weighted imaging (DWI) is a valuable tool for breast cancer detection and supplemental screening in women with dense breasts.
• Current interpretation of breast DWI can be subjective, leading to variability in diagnostic accuracy and potential for unnecessary follow-ups.

Purpose of the Study:

• To evaluate an artificial intelligence (AI)-powered computer-aided diagnosis (CAD) system for its impact on clinical interpretation and workload reduction in breast DWI.
• To assess the AI system's performance in classifying breast lesions based on DWI and its effect on inter-reader agreement.

Main Methods:

• A retrospective study involving 824 examinations for model development and 235 for evaluation.
• Readings were performed by three readers with and without AI-CAD assistance, using BI-RADS-like classification based on DWI.
• The AI model, based on nnDetection, was trained using 5-fold cross-validation and ensembling; performance was assessed using AUC and inter-rater agreement (Cohen's kappa).

Main Results:

• The AI-augmented approach significantly reduced BI-RADS-like 3 calls by 29% (P=.019) and improved inter-rater agreement (0.57 vs 0.49).
• Two readers detected more malignant lesions with AI-CAD assistance.
• The AI model achieved an AUC of 0.78, increasing to 0.82 for women at screening age, indicating potential for 20.9% workload reduction at 96% sensitivity.

Conclusions:

• AI support shows potential to enhance breast DWI interpretation by reducing ambiguous classifications and improving reader consistency.
• The AI-CAD system demonstrated improved diagnostic performance and efficiency, suggesting its utility in clinical practice.
• Further research with larger study cohorts is recommended to validate these findings.