R^2 Net Recurrent and Recursive Network for Sparse View CT Artifacts Removal


We propose a novel neural network architecture to reduce streak artifacts generated in sparse-view 2D Cone Beam Computed To-mography (CBCT) image reconstruction. This architecture decomposes the streak artifacts removal into multiple stages through the recurrent mechanism, which can fully utilize information in previous stages and guide the learning of later stages. In each recurrent stage, the key components of the architecture are computed recursively. The recursive mechanism is helpful to save parameters and enlarge the receptive field effiently with exponentially increased dilation of convolution. To verify its ectiveness, we conduct experiments on the AAPM CBCT dataset through 5-fold cross-validation. Our proposed method outperforms the state-of-the-art methods both quantitatively and qualitatively.

Medical Image Computing and Computer Assisted Interventions