Related Experiment Video
Updated: Jan 7, 2026
11:12
Optimization of a Multiplex RNA-based Expression Assay Using Breast Cancer Archival Material
Published on: August 1, 2018
8.3K
Explainable deep learning for predicting HER-2 expression in breast cancer: a multicenter study
Zhendong Lu1, Minping Hong2, Xinhua Li1
1Department of Radiology, The Affiliated Hospital of Guangdong Medical University, Zhanjiang, PR China.
Acta Radiologica (Stockholm, Sweden : 1987)
|December 30, 2025
Summary
A novel deep learning framework accurately predicts HER-2 status non-invasively using MRI scans. This approach aids in preoperative breast cancer stratification, potentially reducing the need for invasive biopsies.
Area of Science:
- Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Human epidermal growth factor receptor 2 (HER-2) is a critical biomarker for breast cancer prognosis and treatment selection.
- Current HER-2 assessment requires invasive tissue biopsy, posing risks and costs to patients.
Purpose of the Study:
- To create an interpretable deep learning imaging framework for non-invasive prediction of preoperative HER-2 expression in breast cancer.
- To develop a tool that can assist in preoperative patient stratification and treatment planning.
Main Methods:
- Retrospective analysis of MRI data and clinical records from 450 breast cancer patients with confirmed HER-2 status.
- Development of a ResNet-based deep neural network to generate a HER-2 positivity probability score (D-score).
- Integration of the D-score with independent clinical predictors for a combined predictive model, validated using ROC analysis.
Main Results:
- The combined deep learning model achieved an AUC of 0.809 in external validation, surpassing the clinical model.
- Interpretability analysis highlighted the D-score, rim enhancement, and lymph node diameter as key predictors.
- The model demonstrated accurate, non-invasive, and interpretable prediction of HER-2 expression.
Conclusions:
- The proposed deep learning framework offers an accurate and non-invasive method for predicting HER-2 expression in breast cancer.
- This tool can potentially serve as a preoperative stratification method, guiding individualized treatment and reducing reliance on invasive biopsies.
