A Flexible Residual Neural Network for Rician Noise Removal in MR Images

Rician noise removal is a crucial problem for processing magnetic resonance (MR) images. In this paper, we present a flexible residual neural network (FDnResNet) for Rician noise reduction. The proposed model employs noise level map to improve modeling capacity and to adjust the compromise between noise removal and image feature preservation. Additionally, our model uses residual blocks to enhance the information flow between the shallower and deeper layers in the network. The experimental results of FDnResNet on simulated brain MR image database demonstrate the superiority of our proposed residual neural network utilizing noise level map.