JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Development and internal validation of a mammography-based model fusing clinical, radiomics, and deep learning models for sentinel lymph node metastasis prediction in breast cancer

  • 0Department of Radiology, The Second Affiliated Hospital of Harbin Medical University, Harbin, China.
Frontiers in Medicine +

|

September 25, 2025

|

September 25, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

A new mammography-based model combining clinical, radiomics, and deep learning data accurately predicts sentinel lymph node metastasis in breast cancer patients. This fusion model shows promise for improved diagnostic accuracy.

Area Of Science

  • Oncology
  • Medical Imaging
  • Artificial Intelligence in Medicine

Background

  • Sentinel lymph node (SLN) metastasis is a critical prognostic factor in breast cancer.
  • Accurate SLN status evaluation is essential for treatment planning and patient outcomes.
  • Current methods for SLN assessment have limitations, necessitating improved diagnostic tools.

Purpose Of The Study

  • To develop and validate a novel mammography (MG)-based post-fusion model for evaluating SLN status in breast cancer patients.
  • To integrate clinical data, radiomic features, and deep learning models for enhanced predictive performance.
  • To assess the diagnostic accuracy and clinical utility of the proposed fusion model.

Main Methods

  • A cohort of 290 breast cancer patients undergoing MG was used, with data split into training, internal validation, and external test sets.
  • Radiomic (Rad) and deep learning (DL) features were extracted from MG images (MLO and CC views).
  • A post-fusion model (Clinical+Rad+DL) was developed by combining probabilities from single-modal models using support vector machine (SVM), with performance evaluated by AUC, DCA, and calibration curves.

Main Results

  • The Clinical+Rad+DL post-fusion model demonstrated superior discrimination ability compared to single-modal or pre-fusion models.
  • The model achieved an AUC of 0.845 (95% CI: 0.769-0.921) in the internal validation set and 0.825 (95% CI: 0.812-0.932) in the independent test set.
  • Decision curve analysis and calibration curves indicated the model's clinical utility and predictive accuracy.

Conclusions

  • The developed Clinical+Rad+DL post-fusion model effectively predicts SLN metastasis in breast cancer.
  • The probability fusion method proved effective, highlighting the potential of integrating diverse data modalities.
  • This approach shows significant promise for improving the non-invasive prediction of SLN status in breast cancer management.

Keywords: