Breast masses classification using a radiomic analysis in contrast-enhanced spectral mammography

Contrast-Enhancement Spectral Mammography (CESM) is a new mammographic modality that shows a better diagnostic precision compared to digital mammography in dense breasts. In this work, the effect of using radiomic information extracted from CESM studies (images obtained from both low-energy (LM) and recombined (CM), respectively) is evaluated. The proposed strategy fuses radiomic texture features extracted from a selected region of interest (ROI) in both the LM (low-energy ROI) and recombined ROI (CM), respectively. These features are then combined into a single, representative feature vector integrating the information from both ROIs, considered as complementary images. The relevant features are selected by using a Principal Component analysis (PCA). Hence two radiomic information fusion approaches were evaluated, in the first, the extracted features from each ROI were assembled in a single vector and then reduced. In the second approach, the extracted features were first reduced and then combined. Finally, the resulting feature vector is used as input for a binary classifier implemented with a support vector machine (SVM) for classifying breast masses as benign or malignant. The obtained results demonstrated that the combined data effectively represent mammographic mass features, enabling automated differentiation between benign and malignant. These shown an average accuracy rate of 92.3%, with average values of 93.3% sensitivity and 88.8% specificity, with an area under the ROC curve (AUC) of 0.9, respectively.