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 5 不同方法在T2大脑数据上重建结果的评价指标

Tab.5 Evaluation metrics for reconstruction results of different networks on fastMRI-T2Brain

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 40.09 37.36 39.11 36.12 37.54 35.20 37.87 35.88 36.83 36.77 RecurrentVN 40.32 38.06 39.46 36.32 39.11 36.92 37.62 36.07 37.15 34.73 Deep-SLR 36.47 35.26 34.55 32.79 37.72 35.25 37.29 34.45 37.63 36.04 Deepcomplex 38.58 36.02 37.22 32.92 38.27 36.19 38.25 35.14 38.56 36.90 DONet 38.81 36.35 37.47 33.97 38.44 36.26 38.47 35.33 38.61 36.93 SwinMR 36.31 35.16 35.91 34.04 36.91 34.63 37.99 34.31 36.35 31.41 VNTM 40.44 38.22 39.67 36.71 39.99 38.03 39.96 37.03 40.04 38.56 SSIM E2E-VN 0.972 0.959 0.968 0.951 0.964 0.952 0.966 0.952 0.962 0.956 RecurrentVN 0.973 0.962 0.969 0.951 0.968 0.958 0.964 0.952 0.962 0.947 Deep-SLR 0.959 0.946 0.943 0.921 0.963 0.947 0.960 0.938 0.962 0.949 Deepcomplex 0.966 0.950 0.957 0.920 0.964 0.951 0.963 0.942 0.964 0.952 DONet 0.967 0.951 0.959 0.930 0.965 0.952 0.964 0.944 0.964 0.953 SwinMR 0.950 0.940 0.948 0.930 0.952 0.934 0.959 0.932 0.947 0.913 VNTM 0.973 0.962 0.969 0.952 0.971 0.962 0.970 0.955 0.971 0.963