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 4 不同方法在膝盖数据（矢状面质子密度加权序列）上重建结果的评价指标

Tab.4 Evaluation metrics for reconstruction results of different networks on knee data (Sagittal-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 36.64 34.25 36.27 32.52 37.95 35.18 36.68 33.12 36.17 33.49 RecurrentVN 38.83 34.31 36.75 32.83 38.44 35.52 37.47 33.79 37.31 36.40 Deep-SLR 36.27 33.91 34.69 30.41 38.68 36.06 37.92 34.34 38.40 35.98 Deepcomplex 37.67 35.02 35.86 31.06 39.93 37.29 38.97 35.24 39.87 37.78 DONet 38.22 35.75 36.50 32.07 39.95 37.58 39.17 35.63 40.32 38.00 SwinMR 33.99 32.62 33.46 30.81 36.34 33.26 36.93 32.90 35.33 33.79 VNTM 39.83 37.08 38.53 33.33 40.76 38.18 39.90 36.18 41.08 38.85 SSIM E2E-VN 0.940 0.911 0.937 0.875 0.949 0.921 0.942 0.891 0.944 0.913 RecurrentVN 0.954 0.902 0.938 0.875 0.949 0.920 0.939 0.886 0.944 0.926 Deep-SLR 0.930 0.891 0.911 0.819 0.949 0.919 0.942 0.887 0.948 0.916 Deepcomplex 0.944 0.908 0.925 0.837 0.959 0.933 0.951 0.902 0.959 0.937 DONet 0.947 0.916 0.931 0.856 0.959 0.936 0.952 0.908 0.962 0.939 SwinMR 0.899 0.869 0.894 0.830 0.925 0.888 0.925 0.860 0.915 0.882 VNTM 0.959 0.931 0.949 0.881 0.964 0.942 0.957 0.916 0.966 0.946