Multimodal Deep Learning with Routine Clinical Data for Recurrence Risk Stratification in HR+/HER2- Early Breast Cancer - PubMed
10 hours ago
- #Breast Cancer Prognosis
- #Clinical AI Integration
- #Multimodal Deep Learning
- Developed MRRP model integrating routine clinical data for recurrence risk stratification in HR+/HER2- early breast cancer.
- Model uses hierarchical transformer with cross-attention to fuse whole-slide images, ultrasound, and structured clinical parameters.
- Achieved superior performance (C-index = 0.840) with robust AUCs at 3, 5, and 7 years compared to single-modality models.
- Highlighted importance of pathology features and complementary value of ultrasound and clinical data through ablation studies.
- Implemented learnable compensation mechanism for missing modality data to enhance model robustness.
- Provides a practical AI tool for personalized treatment decisions without relying on costly multi-omics data.