Abstract: Background and purpose: Digital breast tomosynthesis (DBT) has been applied to breast cancer screening and diagnosis population, which can improve the breast cancer detection. The purpose of this study was to evaluate the differential diagnosis of breast mass lesions based on DBT images. Methods: In the retrospective study, we analyzed the patients undergoing DBT examination in Fudan University Shanghai Cancer Center between April 2019 and August 2020, who were confirmed by surgery and pathology. Finally, a total of 143 female patients showing mass signs were enrolled in this study. Radiomics features were extracted from 3D images of DBT based on mass lesions, and Lasso logistic regression model was used for feature dimension reduction and screening to establish radiomics labels. The model was built by logistic regression (LR), support vector machine (SVM) and gradient boosting decision tree (GBDT) algorithms. Receiver operating characteristic (ROC) was used to evaluate the diagnostic efficacy of radiomic labels for benign and malignant breast tumors. Results: Among 144 lesions confirmed by pathology, 65 were benign and 79 were malignant. It was divided into training set and test set according to the ratio of 8∶2. Based on the classifier algorithm with different number features, the optimal numbers of features of LR, SVM and GBDT were 20, 24 and 32 respectively. The GBDT model achieved an area under curve (AUC) value of 0.91 on the test set. Conclusion: Due to the advantages of integrated learning, GBDT model based on radiomics could effectively distinguish benign from malignant breast lesions in DBT.

Key words: Digital breast tomosynthesis, Mass, Radiomics, Diagnosis