Predictive value of radiomic signature based on 2-[18F]FDG PET/CT in HER2 status determination for primary breast cancer with equivocal IHC results

Abstract

PurposeTo evaluate the predictive power of 2-[18F]FDG PET/CT-derived radiomic signature in human epidermal growth factor receptor 2 (HER2) status determination for primary breast cancer (BC) with equivocal immunohistochemistry (IHC) results for HER2. MethodsA total of 154 primary BC with equivocal IHC results for HER2 were retrospectively enrolled in the study. First, the following five conventional PET parameters (SUVmax, SUVmean, SUVpeak, MTV, TLG) were measured and compared between HER2-positive and HER2-negative cohorts. After quantitative radiomic features extraction and reduction, the least absolute shrinkage and selection operator (LASSO) algorithm was used to establish a radiomic signature model. Then, the area under the curve (AUCs) after a receiver operator characteristic (ROC) analysis, accuracy, sensitivity and specificity were calculated and used as the main outcomes. Finally, a total of 37 BC patients from an external institution were included to perform an external validation. ResultsAll the five conventional PET parameters were unable to discriminate between HER2-positive and HER2-negative cohorts for BC (P = 0.104–0.544). Whereas, the developed radiomic signature model was potentially predictive of HER2 status with an of AUC 0.887 (95% confidence interval [CI], 0.824–0.950) in the training cohort and 0.766 (95% CI, 0.616–0.916) in the validation cohort, respectively. For external validation, the AUC for the external test cohort was 0.788 (95% CI, 0.633–0.944). ConclusionsRadiomic signature based on 2-[18F]FDG PET/CT images was capable of non-invasively predicting the HER2 status with a comparable ability to FISH assay, especially for those with equivocal IHC results for HER2.