Researchers externally validated three deep learning models—DenseNet121, ViT-B/32, and ResNet50—for predicting breast density from ultrasound images. The models demonstrated strong performance, particularly in extremely dense breasts, though heterogeneously dense breasts remained a challenge. When integrated into a risk prediction model, AI-derived density showed comparable results to mammography-reported density, suggesting generalization across different demographics. AI
Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →
IMPACT Validates generalization of deep learning models for medical imaging tasks, though highlights areas for further optimization in density assessment.
RANK_REASON This is a research paper detailing the external validation of deep learning models for a specific medical imaging task.