Multimodal Deep Learning with Routine Clinical Data for Recurrence Risk Stratification in HR⁺/HER2^- Early Breast...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Hormone receptor-positive/human epidermal growth factor receptor 2-negative (HR+/HER2-) early breast cancer (EBC) patients have a persistent risk of recurrence.
• Existing prognostic tools lack sufficient accuracy and accessibility for precise risk stratification.

Purpose of the Study:

• To develop and validate a multimodal recurrence risk prediction (MRRP) model for HR+/HER2- EBC patients.
• To integrate routinely available clinical data, including whole-slide images (WSIs), ultrasound (US) imaging, and diagnostic reports.

Main Methods:

• Developed a hierarchical transformer-based fusion framework with cross-attention mechanisms to integrate diverse data modalities.
• Utilized a cohort of 768 HR+/HER2- EBC patients with long-term follow-up for model training and validation.
• Implemented a learnable compensation mechanism to handle missing modality data.

Main Results:

• The MRRP model achieved superior prognostic performance (C-index = 0.840) compared to single-modality models.
• Demonstrated robust time-dependent AUCs exceeding 0.85 at 3, 5, and 7 years post-treatment.
• Pathology features were crucial, with complementary value from US and clinical data.

Conclusions:

• The MRRP model provides a clinically practical, AI-driven tool for precise risk stratification in HR+/HER2- EBC.
• Enables individualized treatment and surveillance decisions, reducing reliance on expensive multi-omics data.