Related Experiment Video
Updated: Nov 9, 2025
08:08
Quantification of Breast Cancer Cell Invasiveness Using a Three-dimensional 3D Model
Published on: June 11, 2014
16.0K
Predicting cell invasion in breast tumor microenvironment from radiological imaging phenotypes
Dooman Arefan1, Ryan M Hausler2, Jules H Sumkin1
1Department of Radiology, University of Pittsburgh School of Medicine, 4200 Fifth Ave, Pittsburgh, PA, 15260, USA.
BMC Cancer
|April 8, 2021
Summary
Breast MRI radiomics can predict tumor microenvironment cell types. This study used radiomics and machine learning to link imaging features to cell abundance in breast cancer, showing potential for non-invasive TME assessment.
Area of Science:
- Oncology
- Radiology
- Bioinformatics
Background:
- The tumor microenvironment (TME) composition, including immune and stromal cells, influences cancer progression and treatment response.
- Radiomics, extracting quantitative imaging features, shows promise in predicting clinical outcomes but its ability to predict TME cell infiltration is not well-established.
- Understanding TME cellularity is crucial for personalized cancer therapies.
Purpose of the Study:
- To investigate the predictive capability of radiomic features from Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI) for the abundance of 10 distinct cell types within breast cancer lesions.
- To develop and evaluate machine learning models for predicting cell type infiltration using a radio-genomics approach.
Main Methods:
- A retrospective analysis of 73 breast cancer patients with available imaging and gene expression data from TCIA and TCGA.
- Extraction of 199 radiomic features (shape, texture, kinetic) from DCE-MRI.
- Application of univariate linear regression and multivariate extreme gradient boosting models with Recursive Feature Elimination for feature selection and classification.
- Model performance evaluated using leave-one-out cross-validation and an independent test set, with Area Under the Curve (AUC) as the primary metric.
Main Results:
- Univariate analysis revealed significant associations between specific radiomic features and fibroblast abundance.
- Multivariate models achieved leave-one-out cross-validation AUCs between 0.5 and 0.83 for predicting cell type infiltration.
- Performance on the independent test set showed AUCs ranging from 0.5 to 0.68 for various cell type predictions.
Conclusions:
- Breast MRI-derived radiomics are associated with the abundance of several cell types within the tumor microenvironment across two independent breast cancer cohorts.
- These findings suggest radiomics can non-invasively characterize TME cellularity.
- Larger cohort studies are warranted to further validate and refine these radio-genomic models for clinical application.
