Deep learning-based multi-modal data integration enhancing breast cancer disease-free survival prediction

Area of Science:

• Oncology
• Medical Imaging
• Bioinformatics

Background:

• Breast cancer prognosis remains challenging, necessitating early metastasis detection and precise treatment prediction.
• Developing advanced predictive models is crucial for improving patient outcomes.

Purpose of the Study:

• To develop and validate a novel multi-modal deep learning model for predicting disease-free survival (DFS) in breast cancer patients.
• To integrate preoperative pathology imaging, molecular, and clinical data for enhanced predictive accuracy.

Main Methods:

• Retrospective collection of multi-modal data from The Cancer Genome Atlas and a Chinese institution.
• Development of the Deep Learning Clinical Medicine Based Pathological Gene Multi-modal (DeepClinMed-PGM) model.
• Validation across training (n=741), internal validation (n=184), and external testing (n=95) cohorts.

Main Results:

• The DeepClinMed-PGM model significantly improved DFS prediction accuracy (AUCs up to 0.979 in training, 0.938 in external testing).
• The model demonstrated robust discriminative capabilities across all cohorts, with significant hazard ratios (P < 0.0001).
• High C-index values (0.925, 0.823, 0.864) confirmed the model's predictive performance.

Conclusions:

• The developed DeepClinMed-PGM model offers a promising approach for accurate breast cancer prognosis.
• Integration of multi-modal data enhances predictive power, paving the way for personalized treatment strategies.