Figure 12.

Comparison of bicubic, interpolation [38], advanced methods, and nESRGAN. Among them, (a) is the original high-resolution MRI slice, (e) is the slice from reconstructed MRI in the first reconstruction. (b,f) are new slices after interpolation repair. Besides, (d,g,h) are advanced super-resolution methods in deep learning. (c) is our proposed method nESRGAN(nosed-based enhanced super-resolution generative adversarial network). (b,d,f,h) still exits a few noises and grid lines which missing value lead. (c,g) perform better than others. The performance of visual quality between (c) and (d) is similar. Furthermore, (c) performs better than (d) in the image quality evaluation.