Breast cancer molecular subtype prediction using radiomics signature on synthetic mammography from digital breast tomosynthesis

Abstract

Purpose: To predict molecular subtype of breast cancer using radiomics signature extracted from synthetic mammography reconstructed from digital breast tomosynthesis (DBT). Materials and Methods: From December 2015 to September 2016, 365 patients with pathologically confirmed invasive breast cancer with preoperative DBT were included. Among the 294 patients from December 2015 to July 2016, 150 patients were selected [50 consecutive patients for each molecular subtype (luminal A+B, luminal; HER2-positive, HER2; triple-negative, TN)], and assigned to the training set. A temporally independent validation cohort consisted of 71 patients with breast cancer between August 2016 and September 2016 (50 luminal, 9 HER2, and 12 TN). Total of 129 radiomics features was extracted from the craniocaudal (CC) and mediolateral oblique (MLO) view of the synthetic and digital mammography. In addition, the radiomics features were separately obtained from the marginal and inner portions of the lesion. The performances of binary radiomics classifications for each subtype were measured using the area under the receiver operating characteristic curve (AUC). The radiomics model was built using the elastic-net with ten-fold cross-validation and validated in the independent validation cohort. The clinical model included patient’s age and BI-RADS based image features. Results: In the validation cohort, the radiomics models yielded an AUC of 0.838 for TN, 0.556 for HER2, and 0.645 for luminal subtypes. With the optimal cut-off value of radiomics signature, sensitivity, and specificity of the models in the validation cohort were 83.3% and 79.7% for TN, 11.1% and 79.0% for HER2, 44.0% and 66.7% for luminal subtypes, respectively. There were no significant differences between synthetic mammography and digital mammography in predicting the three subtypes (p = 0.812 for TN, 0.268 for HER2 and 0.833 for luminal). There were no significant differences when comparing the original ROI with the ROIs with only inner and marginal portions (p = 0.084 and 0.051). In multivariate analysis of radiomics signature and clinical features in the classification task of TN versus non-TN, radiomics signature was the only independent predictor for predicting TN subtype. In addition, the combination of radiomics signature and clinical features showed significantly higher AUC value than the clinical features only in distinguishing TN subtype (p = 0.045). Conclusion: The radiomics signature derived from the synthetic mammography from DBT showed high performance in distinguishing between TN and non-TN breast cancer. The radiomics signature from the synthetic 2D mammography of the DBT may serve as a biomarker to distinguish TN subtype of breast cancer and may affect the direction of treatment.