Digital breast tomosynthesis versus digital mammography: integration of image modalities enhances deep learning-based breast mass classification

Objective To evaluate the impact of utilizing digital breast tomosynthesis (DBT) or/and full-field digital mammography (FFDM), and different transfer learning strategies on deep convolutional neural network (DCNN)-based mass classification for breast cancer. Methods We retrospectively collected 441 patients with both DBT and FFDM on which regions of interest (ROIs) covering the malignant, benign and normal tissues were extracted for DCNN training and validation. Experiments were conducted for tasks in distinguishing malignant/benign/normal: (1) classification capabilities of DBT vs FFDM and the role of transfer learning were validated on 2D-DCNN; (2) different strategies of combining DBT and FFDM and the associated impacts on classification were explored; (3) 2D-DCNN and 3D-DCNN trained from scratch with volumetric DBT were compared. Results 2D-DCNN with transfer learning outperformed that without for DBT in distinguishing malignant (Delta AUC = 0.059 +/- 0.009, p < 0.001), benign (Delta AUC = 0.095 +/- 0.010, p < 0.001) and normal tissue (Delta AUC = 0.042 +/- 0.004, p < 0.001) (paired samples t test). 2D-DCNN trained on DBT (with transfer learning) achieved higher accuracy than those on FFDM (malignant: Delta AUC = 0.014 +/- 0.014, p = 0.037; benign: Delta AUC = 0.031 +/- 0.006, p < 0.001; normal: Delta AUC = 0.017 +/- 0.004, p < 0.001) (independent samples t test). The 2D-DCNN employing both DBT and FFDM for training achieved better performances in benign (FFDM: Delta AUC = 0.010 +/- 0.008, p < 0.001; DBT: Delta AUC = 0.009 +/- 0.005, p < 0.001) and normal (FFDM: Delta AUC = 0.005 +/- 0.003, p < 0.001; DBT: Delta AUC = 0.002 +/- 0.002, p < 0.001) (related samples Friedman test). The 3D-DCNN and 2D-DCNN trained from scratch with DBT only produced moderate classification. Conclusions Transfer learning facilitates mass classification for both DBT and FFDM, and DBT outperforms FFDM when equipped with transfer learning. Integrating DBT and FFDM in DCNN training enhances mass classification accuracy for breast cancer.