MRI radiomics model predicts breast cancer immune marker before surgery

Doctors may soon be able to estimate how strongly a breast tumor is infiltrated by immune cells before a patient ever enters the operating room. A retrospective two-center study published in BMC Medical Imaging reports that a model combining multiphase contrast-enhanced MRI radiomics with conventional preoperative features predicted tumor-infiltrating lymphocyte (TIL) levels with an area under the curve of 0.882 (95% CI: 0.840–0.920) in the training set, 0.777 (95% CI: 0.672–0.866) in internal validation, and 0.738 (95% CI: 0.637–0.826) in external validation.

TILs are biomarkers of the breast cancer immune microenvironment that are associated with treatment response and prognosis, but their assessment currently relies mainly on surgical histopathology, meaning the information arrives only after the tumor has been removed. The study included 495 patients who underwent preoperative breast DCE-MRI followed by surgical resection as their initial treatment. TILs were assessed on surgical specimens using a prespecified 10% cutoff commonly adopted in previous breast MRI-based TIL studies.

What distinguishes the new approach is its attention to tissue surrounding the tumor as well as the tumor itself. The researchers extracted radiomic features from tumoral regions and a 3-millimeter peritumoral rim across six DCE-MRI phases, then combined these with clinical, semantic MRI, and biopsy-derived features. Performance was evaluated through discrimination, calibration, incremental value, and clinical utility, and Shapley additive explanations (SHAP) analysis was used to determine which features drove the predictions.

The combined model outperformed simpler alternatives. In external validation, it showed a significant integrated discrimination improvement over the conventional model (IDI = 0.105, P = 0.005) and over the tumoral–peritumoral model (IDI = 0.098, P = 0.001). Decision curve analysis showed greater net benefit across most evaluated threshold probabilities, suggesting the model could offer clinical value across a range of decision settings.

SHAP analysis identified later-phase peritumoral radiomic features as the dominant predictive factors for TIL estimation. This finding points to the immune-relevant information contained in the tissue immediately surrounding the tumor, particularly in later phases of contrast enhancement, as a key signal the model learned to read.

The authors conclude that integrating multiphase tumoral–peritumoral DCE-MRI radiomics with conventional preoperative features showed potential for predicting TIL levels, with late-phase peritumoral features dominating the predictive performance. They describe the model as a promising preoperative tool for immune-related risk stratification and individualized multidisciplinary treatment planning.

As a retrospective study, the work establishes predictive associations rather than demonstrating a direct clinical benefit, and the reported performance figures come from the study’s own validation sets. The authors note the article is being shared early as peer-reviewed, accepted research carrying a permanent DOI, subject to further edits before the final version of record.

Subject of Research: Preoperative prediction of breast cancer tumor-infiltrating lymphocyte levels using DCE-MRI radiomics

Article Title: Preoperative prediction of breast cancer tumor-infiltrating lymphocytes levels with multiphase tumoral–peritumoral DCE-MRI radiomics and conventional features

Article References: Yang, W., Chen, M., Cui, Y., Xu, W., Tian, W., Jiang, Y., Xu, J., Cai, Y., Gu, Y., & Peng, X.-G. (2026). Preoperative prediction of breast cancer tumor-infiltrating lymphocytes levels with multiphase tumoral–peritumoral DCE-MRI radiomics and conventional features. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02930-y

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02930-y

Keywords: breast cancer, tumor-infiltrating lymphocytes, DCE-MRI, radiomics, peritumoral features, immune microenvironment, preoperative prediction, machine learning, BMC Medical Imaging, SHAP analysis, Preoperative, prediction