T2WI-based radiomics signatures for predicting human epidermal growth factor receptor 2 status in breast cancer / 中国医学影像技术

ABSTRACT

Objective To investigate the value of T2WI-based radiomics signatures for preoperatively prediction of human epidermal growth factor receptor 2 (HER2) status in breast cancer. Methods T2WI of 209 patients with breast cancer were retrospectively analyzed. The patients were randomly divided into training group (n=145) and validation group (n=64). The ROIs were manually delineated around the tumor profile. Radiomics feature extraction was implemented in MATLAB 2013a. The interclass correlation coefficients, the least absolute shrinkage and selection operator Logistic regression analysis were used for radiomics features selection and generation. The difference of the Rad-score between HER2-positive and HER2- negative subgroups was observed. The predictive performances of the radiomics signatures for HER2 status were evaluated with ROC curves in training group, and were validated in validation group with the obtained predictive threshold. Results The radiomics signatures were constituted by 13 selective features. In both training and validation groups, there were statistically significant differences in Rad-score between HER2-positive subgroup and HER2-negative subgroup (both P<0.05). T2WI-based radiomics signatures exhibited good discrimination for HER2 status, with AUC of 0.798 in training group and 0.707 in validation group. Conclusion The radiomics signatures based on T2WI have certain value for preoperative prediction of HER2 status in breast cancer.