1: \begin{abstract}

2: Obtaining accurate and reliable images from low-dose computed tomography (CT) is challenging.

3: Regression convolutional neural network (CNN) models that are learned from training data are increasingly gaining attention in low-dose CT reconstruction.

4: This paper modifies the architecture of an iterative regression CNN, \emph{BCD-Net}, for fast, stable, and accurate low-dose CT reconstruction, and presents the convergence property of the modified BCD-Net.

5: Numerical results with phantom data show that applying faster numerical solvers to model-based image reconstruction (MBIR) modules of BCD-Net leads to faster and more accurate BCD-Net; 

6: BCD-Net significantly improves the reconstruction accuracy, compared to the state-of-the-art MBIR method using learned transforms; 

7: BCD-Net achieves better image quality, compared to a state-of-the-art iterative NN architecture, ADMM-Net.

8: Numerical results with clinical data show that BCD-Net generalizes significantly better than a state-of-the-art deep (non-iterative) regression NN, FBPConvNet, that lacks MBIR modules.

9: \blfootnote{The authors indicated by asterisks ($^\ast$) equally contributed to this work.}

10: \blfootnote{Corresponding author: Yong Long (email: \email{yong.long@sjtu.edu.cn}).}

11: \blfootnote{This paper has supplementary document. The prefix \dquotes{S} indicates the numbers in figure and section in the supplementary document.}

12: \end{abstract}

13: