Figure 3.

Comparisons to the residual encoder–decoder convolutional neural network (RED-CNN). The proposed MS-RDN was compared with RED-CNN in three sets of configurations: single slice training without TOI oriented patch extraction ($Z=1$, nonTOI), single slice training ($Z=1$), and multi-slice training ($Z=5$). Breast images of the test subject were reconstructed by these RED-CNNs and MS-RDNs using the retrospectively undersampled 100-view data. The reference images were obtained using FDK on the 300-view data. The display window is $[0.15,0.35]{\text{cm}}^{-1}$.