Metal Artifact Reduction for CT Method Based on Attention Gate UNet
Currently,the basic UNet model cannot effectively meet the demand for CT images with metal artifact reduction,the simple structure of the UNet cannot precisely extract the accurate information on the effective structure and details,and deep convolu-tional neural network does not sufficiently use the information of low-level features.Based on the above problems,a metal artifact re-moval network with attention gates based on the UNet is proposed.The network adopts the attention gates to process the attention weights of the information at low and high levels,The jump connection mechanism and feature decoding structure are used to improve the quality of the generated CT images,the final CT images with metal artifact reduction are obtained through the multilevel encoding and decoding structure.The experimental results show that the proposed method achieves better visual effects in removing stripe and banding artifacts in CT images,with a PSNR of 35.591 3,FSIM of 0.961 3,and SSIM of 0.928 8.Compared to existing methods such as the ADN,cGANMAR,UNet,CNNMAR,and CycleGAN,the proposed method has significant advantages in multiple aspects.
metal artifact reductionattention mechanismconvolutional neural networkUNetencoder-decoder
万方数据
维普
