Optimized prediction of breast cancer tumor microenvironment using MRI-based intratumoral and peritumoral radiomics: a prospective study

TL;DR

This study shows that combining MRI scans of tumor and surrounding areas can predict the tumor microenvironment in breast cancer, potentially guiding personalized treatments.

Contribution

The study introduces a novel approach using combined intratumoral and peritumoral MRI radiomics to predict tumor microenvironment components in breast cancer.

Findings

Combined intratumoral and peritumoral radiomic features outperformed intratumoral-only features in predicting TME components.

Contrast-enhanced T1-weighted MRI sequences provided the highest predictive accuracy for ECM and immune cell abundance.

Combined models better captured ECM-rich and immunosuppressive TME profiles linked to poor prognosis.

Abstract

The tumor microenvironment (TME), composed of non-tumor elements such as stromal matrix and immune cells, plays a critical role in tumor progression, metastasis, and treatment response. This study aimed to investigate the association between MRI-based intratumoral and peritumoral radiomic features and the TME components, including extracellular matrix (ECM) and immune cells, in patients with invasive breast cancer.

In this prospective study, 121 women with histologically confirmed invasive breast cancer underwent pre-treatment multiparametric 3T breast MRI, including T2-weighted, diffusion-weighted imaging (DWI), and dynamic contrast-enhanced T1-weighted sequences (NCT06095414, registered at ClinicalTrials.gov). The dataset was randomly divided into training and testing cohorts in a 7:3 ratio. A total of 16180 radiomic features were extracted from both intratumoral and peritumoral regions. Three-dimensional volume histology with quantitative immunohistochemical staining of ECM and immune cells served as the reference standard for TME assessment. Predictive models were developed using least absolute shrinkage and selection operator regression and evaluated using area under the receiver-operating characteristic curve (AUC). Model performance was compared between intratumoral-only and combined intratumoral–peritumoral features across five MRI sequences.

Models incorporating both intratumoral and peritumoral features significantly outperformed those using intratumoral features alone in predicting TME components (P < 0.01). Among the five sequences, initial and delayed postcontrast T1-weighted images yielded the highest AUCs. For ECM abundance, the AUCs (95% CI) were 0.82 (0.78–0.87) and 0.82 (0.78–0.88) on initial and delayed imaging, respectively. For immune cell abundance, the AUCs were 0.82 (0.77–0.87) and 0.83 (0.78–0.88). Most of the top predictive features were first-order and texture features associated with tissue heterogeneity. Combined models more accurately captured ECM-rich and immunosuppressive TME profiles, characterized by elevated regulatory T cells and reduced cytotoxic T cells, which were associated with poor prognosis.

MRI-based radiomic features from both intratumoral and peritumoral regions are significantly associated with TME components in invasive breast cancer. Contrast-enhanced T1-weighted sequences provided the most robust performance. These findings highlight the potential of MRI-based radiomics as a powerful noninvasive biomarker for characterizing the TME and informing personalized therapeutic strategies, including immunotherapy and ECM-targeted treatments.

Figures