基于变分模型和Transformer的多尺度并行磁共振成像重建

段继忠,李海源

Multi-scale parallel magnetic resonance imaging reconstruction based on variational model and Transformer

Jizhong DUAN,Haiyuan LI

表 3 不同方法在膝盖数据（冠状面质子密度加权序列）上重建结果的评价指标

Tab.3 Evaluation metrics for reconstruction results of different methods on knee data (Coronal-PD)

评价指标 方法 3× 1DRU 5× 1DRU 3× 1DUU 5× 1DUU 5× 2DRU 10× 2DRU 5× RADU 10× RADU 5× 2DPU 10× 2DPU PSNR/dB E2E-VN 36.39 34.75 34.33 32.26 36.27 33.13 35.86 31.82 34.64 32.11 RecurrentVN 37.61 35.46 35.17 32.71 36.44 33.00 36.08 33.95 35.75 34.00 Deep-SLR 33.75 32.22 31.05 28.74 36.68 34.23 36.64 33.20 35.96 34.29 Deepcomplex 36.55 33.95 33.43 30.13 38.73 36.19 38.29 34.79 38.82 36.64 DONet 36.85 34.58 34.08 31.31 39.02 36.37 38.52 35.08 39.34 36.88 SwinMR 33.95 33.10 32.89 31.99 35.86 32.85 36.67 32.83 35.13 33.65 VNTM 37.79 35.80 36.36 33.57 39.22 36.76 38.91 35.52 39.76 37.37 SSIM E2E-VN 0.938 0.916 0.921 0.886 0.940 0.904 0.934 0.880 0.933 0.898 RecurrentVN 0.949 0.920 0.926 0.885 0.937 0.899 0.931 0.896 0.940 0.907 Deep-SLR 0.902 0.868 0.874 0.811 0.938 0.905 0.937 0.882 0.935 0.902 Deepcomplex 0.938 0.898 0.899 0.839 0.953 0.923 0.949 0.902 0.956 0.929 DONet 0.941 0.908 0.910 0.861 0.955 0.925 0.950 0.906 0.959 0.932 SwinMR 0908 0.886 0.893 0.868 0.923 0.884 0.927 0.866 0.919 0.887 VNTM 0.949 0.921 0.935 0.893 0.956 0.930 0.952 0.911 0.962 0.937