Research Article

A Network with Composite Loss and Parameter-free Chunking Fusion Block for Super-Resolution MR Image

Table 1

The super-resolution reconstructed LR image is obtained by bicubic downsampling (BD). The experimental data show the comparison of other methods with our proposed method in various evaluation indexes: PSNR, SSIM, and RMSE.

Bicubic downsampling (BD)
PDT1T2Params (MB)
PSNRSSIMRMSEPSNRSSIMRMSEPSNRSSIMRMSE

Bicubic32.630.96750.024332.100.95420.025632.710.96010.0236
ESPCNN [42]32.070.96470.025931.520.95140.027132.030.95550.02530.22
VRCNN [48]33.670.97430.021732.840.96320.023634.370.97130.01960.21
EDSR [25]34.400.97820.020433.030.96510.023135.510.97560.01735.08
PAN [49]34.330.97830.020333.100.96520.023035.370.97590.01760.72
NAFNet [50]32.530.96760.024831.970.95510.026332.630.95940.023916.58
Proposed34.520.97900.019933.230.96590.022735.530.97610.01735.27

Bold values highlight the optimal results.