# MRI-based radiomics nomogram for predicting CD8-positive tumor-infiltrating lymphocytes levels in HER2-positive breast cancer

## TL;DR

This study creates a model using MRI scans and clinical data to predict immune cell levels in HER2-positive breast cancer patients, which could help identify those who may benefit from immunotherapy.

## Contribution

A novel MRI-based radiomics nomogram is developed to predict CD8+ TILs levels in HER2-positive breast cancer patients.

## Key findings

- The radiomics nomogram achieved an AUC of 0.866 in the training cohort and 0.886 in the validation cohort.
- The model outperformed the clinical-imaging model and showed comparable performance to the radiomics signature alone.
- The model combines radiomic features with clinical-imaging factors like tumor margin and enhancement pattern.

## Abstract

To develop a radiomics nomogram based on radiomic features derived from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) combined with clinical-imaging characteristics in predicting the CD8+Tumor-infiltrating lymphocytes (TILs) levels in patients with human epidermal growth factor receptor 2 (HER2)-positive breast cancer (BC).

A total of 126 BC patients with pathologically confirmed HER2-positive were enrolled and randomly divided into training (n = 88) and validation (n = 38) cohorts. A clinical-imaging model was built based on clinical and MRI characteristics. Radiomics features were extracted from the third post-contrast phase on DCE-MRI. Select K Best, the maximum relevance minimum redundancy (mRMR), and least absolute shrinkage and selection operator algorithm (LASSO) were used to select radiomics features and a radiomics signature score (rad-score) was constructed by seven radiomics features. Multivariate logistic regression analysis was used to construct a radiomics nomogram model by combining with rad-score and independent clinical-imaging factors. Performance of the clinical-imaging model, rad-score, and radiomics nomogram model were evaluated using the area under the curve (AUC).

Seven radiomics features were used to build the rad-score. The rad-score achieved good performance in predicting CD8+TILs with AUCs= 0.853 and 0.822, respectively. The radiomics nomogram model based on rad-score and clinical-imaging features (tumor margin and enhancement pattern) yielded an optimal AUC of 0.866 and 0.886 in the training and validation cohorts, respectively. The radiomics nomogram significantly outperformed the clinical-imaging model (p < 0.05) and showed a trend toward better performance compared to the rad-score alone (p > 0.05).

The MRI-based radiomics nomogram has the ability to predict CD8+TILs levels, which could be useful in identifying potential in HER2-positive BC patients who can benefit from immunotherapy.