Randomized trial evaluates a dual-view mammography framework for improved accuracy in breast cancer detection, suggesting its utility in clinical practice.
Key Points
This research aims to develop a trustworthy dual-view mammography framework for breast cancer diagnosis, addressing existing challenges in accuracy.
Utilized a Physics-Aware Multi-scale Feature Cooperative Embedding mechanism for data interpretation.
Implemented a Density-Adaptive Feature Discrimination module to manage glandular density in images.
Adopted an Interactive Graph Attention fusion architecture for soft alignment of images.
Achieved an accuracy of 90.10% with AUC = 0.9450 on the Chinese Mammography Database.
Significantly outperformed state-of-the-art methods across multiple key metrics.
Validated the alignment of model decisions with radiological priors, enhancing trust in AI diagnostics.