Table 10.
Performance of hybrid deep learning methods with and without the attention layers in their k-space deep learning components on the Stanford dataset.
| Methods | NMSE (× 103) ↓ | PSNR ↑ | SSIM (× 10∈2) ↑ | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Cartesian | Radial | Cartesian | Radial | Cartesian | Radial | |||||||
| mean±std | p-value | mean±std | p-value | mean±std | p-value | mean±std | p-value | mean±std | p-value | mean±std | p-value | |
| KIKI-Net | 10.96±0.01 | 2.1e-23 | 10.44±0.01 | 9.6e-22 | 35.71±1.75 | 5.5e-32 | 35.91±1.72 | 2.5e-39 | 91.98±0.02 | 9.7e-24 | 92.78±0.02 | 4.1e-24 |
| w/attention | 10.84±0.01 | - | 10.08±0.01 | - | 35.76±1.74 | - | 36.07±1.76 | - | 92.00±0.02 | - | 92.80±0.02 | - |
| Hybrid-Net | 10.77±0.01 | 9.5e-22 | 10.39±0.01 | 7.4e-17 | 35.79±1.63 | 4.3e-32 | 35.95±1.67 | 1.7e-34 | 91.98±0.02 | 3.5e-21 | 92.81±0.02 | 1.6e-19 |
| w/attention | 10.40±0.01 | - | 10.01±0.01 | - | 35.95±1.59 | - | 36.11±1.70 | - | 92.13±0.02 | - | 92.84±0.02 | - |