CEM Radiomics for Distinguishing Benign vs Malignant Lesions in Patients with Invasive Breast Cancer or Benign Breast Lesions

In this retrospective study, 126 women with 143 biopsy proven invasive breast cancers (IBC) or benign lesions were imaged using contrast-enhanced mammography (CEM), prior to any treatment. Size-matched segmentations were manually contoured by an experienced breast radiologist on the CEM capturing the breast lesion (denoted as ROI), breast parenchymal enhancement (BPE), and the fat, respectively. Radiomics analysis was performed using LIFEx software and 111 radiomics metrics spanning 6 different texture families were extracted from each segmentation. Predictive models of benign vs malignant lesions were developed using CEM radiomics models Random Forest, Real Adaboost, and ElasticNet classifiers with data from the ROI only, BPE only, and ROI + BPE. Radiomics values extracted from fat region were then subtracted from the radiomics metrics extracted from the ROI and BPE, respectively, to enhance data harmonization. The 10- fold cross validation was used to assess model performance. RF classifier achieved an AUC of 0.92 for the ROI + BPE. The Adaboost classifier achieved an AUC of 0.97 for the ROI only, and 0.90 for the BPE only. The high performance across all three assessments showed that CEM radiomics can distinguish between benign and malignant breast lesions using information from the background parenchyma with and without the lesion itself.