MRI-based multimodal AI model enables prediction of recurrence risk and adjuvant therapy in breast cancer

Abstract

Timely intervention and improved prognosis for breast cancer patients rely on early metastasis risk detection and accurate treatment predictions. This study introduces an advanced multimodal MRI and AI-driven 3D deep learning model, termed the 3D-MMR-model, designed to predict recurrence risk in non-metastatic breast cancer patients. We conducted a multicenter study involving 1,199 non-metastatic breast cancer patients from four institutions in China, with comprehensive MRI and clinical data retrospectively collected. Our model employed multimodal-data fusion, utilizing contrast-enhanced T1-weighted imaging (T1+C) and T2-weighted imaging (T2WI) volumes, processed through a modified 3D-UNet for tumor segmentation and a DenseNet121-based architecture for disease-free survival (DFS) prediction. Additionally, we performed RNA-seq analysis to delve further into the relationship between concentrated hotspots within the tumor region and the tumor microenvironment. The 3D-MR-model demonstrated superior predictive performance, with time-dependent ROC analysis yielding AUC values of 0.90, 0.89, and 0.88 for 2-, 3-, and 4-year DFS predictions, respectively, in the training cohort. External validation cohorts corroborated these findings, highlighting the model’s robustness across diverse clinical settings. Integration of clinicopathological features further enhanced the model’s accuracy, with a multimodal approach significantly improving risk stratification and decision-making in clinical practice. Visualization techniques provided insights into the decision-making process, correlating predictions with tumor microenvironment characteristics. In summary, the 3D-MMR-model represents a significant advancement in breast cancer prognosis, combining cutting-edge AI technology with multimodal imaging to deliver precise and clinically relevant predictions of recurrence risk. This innovative approach holds promise for enhancing patient outcomes and guiding individualized treatment plans in breast cancer care. • A multi-task 3D deep learning MRI-based multimodal data model (3D-MMR-model) was developed for tumor segmentation and disease-free survival (DFS) prediction. An innovative multimodal approach integrating clinicopathological data with deep learning MRI insights yielded the 3D-MMR-model, surpassing standalone MRI models in predictive precision. Visualization techniques provided insight into decision-making processes, correlating model predictions with the tumor microenvironment.