MRI Radiomics and Vascular Gene Profiles Capture Different Sides of Breast Cancer Tumor Blood Vessels

Tumor blood vessels are more than passive plumbing. They shape how chemotherapy reaches cancer cells, how oxygen and nutrients are delivered, and how a tumor behaves under treatment pressure. A new study in Breast Cancer Research argues that the two main ways clinicians and scientists can read this vascular network — dynamic contrast-enhanced MRI and vascular-related gene expression — do not simply duplicate each other. Instead, each captures a different slice of the same biological reality, and combining them may offer a more complete picture of which patients respond to treatment.

The research team, led by Qingpei Lai and Ge Ren with colleagues including Jing Cai, analyzed paired pre-treatment DCE-MRI scans and mRNA expression data from 932 patients with high-risk stage II or III breast cancer. DCE-MRI tracks how contrast agent flows through tissue over time, offering a macroscopic view of tumor perfusion. Gene expression profiling, by contrast, reveals what the vascular biology looks like at the molecular level. The study asked whether the subgroups defined by each modality overlap, and whether they carry distinct clinical and biological signals.

To answer this, the researchers identified imaging subgroups using k-means clustering of repeatable radiomic features extracted from the MRI scans, and gene subgroups using k-means clustering of vascular-related genes. The procedure produced three imaging subgroups, accounting for 24.8 percent, 34.3 percent, and 40.9 percent of patients, and three gene subgroups, accounting for 33.6 percent, 29.2 percent, and 37.2 percent. If both approaches were reading the same underlying vascular phenotype, these subgroups should have aligned closely. They did not.

The statistical association between imaging and gene subgroups was weak, with a Cramér’s V of 0.068 and a p-value of 0.073, which did not reach conventional significance. This near-independence is the study’s central finding. It suggests that the macroscopic vascular architecture visible on MRI and the microscopic gene-expression programs active within the tumor are not redundant measurements but complementary dimensions of vascular heterogeneity.

The two subgroup systems also related differently to established breast cancer classifications. Imaging subgroups showed weak associations with both receptor-defined subtypes and PAM50 molecular subtypes, with Cramér’s V values of 0.073 and 0.115 respectively. Gene subgroups showed considerably stronger associations, at 0.306 and 0.418. In other words, the gene-based vascular clusters track more closely with known molecular biology, while the imaging-based clusters appear to capture vascular properties that are not well explained by receptor status or PAM50 classification alone.

Pathway enrichment analysis reinforced this separation. Imaging subgroups were associated with metabolic pathways and stromal signaling, while gene subgroups were linked to DNA repair, oxidative metabolism, and immune signaling, all with false discovery rate q-values below 0.05. These distinct enrichment patterns indicate that each modality is picking up different biological programs operating within the tumor microenvironment, rather than converging on the same pathways from different angles.

The clinical implications emerged when the researchers examined pathological complete response, or pCR, after neoadjuvant treatment, adjusting for hormone receptor and HER2 status. Image subgroup 2 was associated with higher pCR, with an adjusted odds ratio of 1.409 and a likelihood ratio p-value of 0.027. Gene subgroup 1 was associated with lower pCR, with an adjusted odds ratio of 0.597, while gene subgroup 2 was associated with higher pCR, with an adjusted odds ratio of 2.217 and likelihood ratio p-values below 0.003. The two systems thus flag different patients as likely responders, and neither fully substitutes for the other.

Prognostic discrimination showed a similar pattern of partial overlap. When the researchers evaluated distant recurrence-free survival, imaging subgroups achieved a C-index of 0.624 compared with 0.589 for gene subgroups, with overall log-rank p-values of 0.008 and 0.015 respectively. Imaging subgroups offered somewhat stronger prognostic separation in this cohort, but both modalities contributed independent information about recurrence risk.

The authors conclude that DCE-MRI radiomic subgroups and vascular genomic subgroups capture complementary dimensions of tumor vascular heterogeneity, and that their distinct biological and clinical associations support the potential value of integrating imaging- and gene-based vascular phenotypes for patient stratification in breast cancer. Because the analysis was conducted in a single cohort of high-risk stage II and III patients and relied on pre-treatment imaging and expression data, validation in independent populations will be needed before such integration can inform routine clinical decision-making. The study is published as open access research in Breast Cancer Research.

Subject of Research: Complementary characterization of tumor vascular heterogeneity in breast cancer using DCE-MRI radiomics and vascular genomics

Article Title: Integrative analysis of DCE-MRI radiomics and genomics reveals complementary characterization of treatment response–associated tumor vascular heterogeneity in breast cancer

Article References: Lai, Q., Teng, X., Zhang, X., Zhang, J., Jiang, Y., Pu, Y., Yu, P., Cai, J., & Ren, G. (2026). Integrative analysis of DCE-MRI radiomics and genomics reveals complementary characterization of treatment response–associated tumor vascular heterogeneity in breast cancer. Breast Cancer Research. https://doi.org/10.1186/s13058-026-02422-5

Image Credits: AI Generated

DOI: 10.1186/s13058-026-02422-5

Keywords: breast cancer, DCE-MRI, radiomics, tumor vasculature, gene expression, pathological complete response, neoadjuvant therapy, tumor heterogeneity, PAM50, distant recurrence-free survival, k-means clustering, molecular subtypes