RADIFUSION: a multi-radiomics deep learning based breast cancer risk prediction model using sequential mammographic images with image attention and bilateral asymmetry refinement.

This study presents RADIFUSION, a novel deep learning model designed for breast cancer risk assessment using sequential screening mammograms.

The model integrates several key features to enhance prediction accuracy:

• Linear image attention mechanism
• Radiomics features
• Gating system to combine different mammographic views
• Bilateral asymmetry-based fine-tuning

Tested on the Cohort of Screen-Aged Women dataset (8,723 patients), RADIFUSION outperformed other state-of-the-art methods, achieving AUCs of:

• 0.905 at 1-year risk interval
• 0.872 at 2-year risk interval
• 0.866 at 3-year risk interval

The results highlight the value of incorporating spatiotemporal data from sequential mammograms and advanced deep learning techniques for improved breast cancer risk evaluation.

This approach offers a promising clinical tool for early detection and patient triage.