# Multiparametric MRI radiomics for noninvasive prediction of HER2 status in immunohistochemical 2+ breast cancer

## TL;DR

This study explores using MRI radiomics to predict HER2 status in breast cancer patients with IHC 2+ results, potentially avoiding invasive tests.

## Contribution

The novel contribution is the development of a noninvasive radiomics model using mpMRI to predict HER2 status in IHC 2+ breast cancer patients.

## Key findings

- The combined mpMRI model achieved an AUC of 0.952 in the training set and 0.892 in the testing set.
- DCEphase7-based radiomics models showed the best single-sequence performance with AUCs of 0.874 and 0.835.
- The clinical-imaging model had lower predictive performance with AUCs of 0.733 and 0.738.

## Abstract

Breast cancer (BC) patients with low expression of human epidermal growth factor receptor 2 (HER2), now classified as HER2-low, may benefit from new novel antibody-drug conjugates, such as trastuzumab deruxtecan. However, patients with immunohistochemistry (IHC) 2+ require further fluorescence in situ hybridization (FISH) to definitively determine HER2 status. Our study aims to explore the feasibility of radiomics models based on multiparametric magnetic resonance imaging (mpMRI) in predicting the HER2 status of IHC 2+ BC patients.

107 IHC 2+ BC patients, which composed of 65 (60.75%) FISH-negative and 42 (39.25%) FISH-positive, were enrolled in the retrospective study and divided into the training set (n = 74) and testing set (n = 33). The clinical and conventional MRI characteristics were used to build a clinical-imaging model. Radiomics features were extracted from T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and DCEphase3 and DCEphase7 in the dynamic contrast-enhanced T1-weighted imaging (DCE-T1WI), and radiomics models were built based on each individual sequence, mpMRI sequences, and a combination of mpMRI sequences with clinical-imaging characteristics. The area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy were calculated. DeLong test was used to determine the differences among various models. p-value <0.05 was considered statistically significant.

The clinical-imaging model achieved AUCs of 0.733 and 0.738 in the training and testing sets. Among the single-sequence radiomics models, the DCEphase7 model achieved the best predictive performance, with AUCs of 0.874 and 0.835 in the training and testing sets. The combined model C, composed of mpMRI model C (DCEphase7+DWI+T2WI) and clinical-imaging model, showed the best predictive performance, with AUCs of 0.952 and 0.892 in the training and testing sets.

The mpMRI-based radiomics models had potential to noninvasively predict the HER2 status in IHC 2+ BC patients.