Related Experiment Video
Updated: Sep 19, 2025
11:12
Optimization of a Multiplex RNA-based Expression Assay Using Breast Cancer Archival Material
Published on: August 1, 2018
8.1K
A Dynamic Contrast-Enhanced MRI-Based Vision Transformer Model for Distinguishing HER2-Zero, -Low, and -Positive
Xu Zhang1,2,3, Yi-Yuan Shen1,3, Guan-Hua Su2,3
1Department of Radiology, Fudan University Shanghai Cancer Center, Fudan University, Shanghai, 200032, China.
Summary
A new Vision Transformer (ViT) model using MRI images accurately predicts human epidermal growth factor receptor 2 (HER2) expression in breast cancer patients. This non-invasive method aids in personalized treatment strategies for HER2-zero, -low, and -positive tumors.
Area of Science:
- Oncology
- Radiology
- Artificial Intelligence
Background:
- Novel antibody-drug conjugates show promise for breast cancer patients with low human epidermal growth factor receptor 2 (HER2) expression.
- Accurate HER2 expression assessment is crucial for guiding targeted therapies and improving patient outcomes.
Purpose of the Study:
- To develop and validate a Vision Transformer (ViT) model using dynamic contrast-enhanced MRI (DCE-MRI) for classifying HER2-zero, -low, and -positive breast cancer.
- To explore the interpretability of the ViT model and its correlation with biological pathways and patient survival.
Main Methods:
- A ViT model was trained and validated on DCE-MRI data from 708 patients (FUSCC cohort) and tested on external cohorts (GFPH and FHCMU).
- The model analyzed early enhancement patterns in MRI images to predict HER2 expression levels.
- Transcriptomics and Cox regression analyses were performed to investigate biological differences and prognostic significance.
Main Results:
- The ViT model achieved significant AUCs (0.71-0.80) in distinguishing HER2-zero from HER2-low/positive tumors across cohorts.
- High AUCs (0.79-0.86) were obtained for classifying HER2-low versus HER2-positive cases.
- The model's prediction score was an independent prognostic factor for overall survival (HR=2.52, p=0.007), and transcriptomics revealed immune-related pathway differences.
Conclusions:
- The developed ViT model offers a non-invasive, accurate method for predicting HER2 expression in breast cancer.
- This approach facilitates precise patient stratification for personalized treatment strategies.
- Further prospective studies are needed to confirm clinical utility.
