Table 5 Comparison of MR-IPT and IPT performance. MR-IPT-level is used as the baseline for evaluation, with both zero-shot and fine-tuned results for various undersampling scenarios across different sampling masks.
From: Magnetic resonance image processing transformer for general accelerated image restoration
| Â | Knee-Cartesian Random | Brain-Cartesian Equispaced | ||||||
|---|---|---|---|---|---|---|---|---|
|  | ACC = 4X | ACC = 8X | ACC = 4X | ACC = 8X | ||||
| Â | PSNR | SSIM | PSNR | SSIM | PSNR | SSIM | PSNR | SSIM |
| Â | Zero-shot | |||||||
 MR-IPT | 34.22 | 0.8635 | 30.65 | 0.7814 | 39.90 | 0.9756 | 32.52 | 0.9288 |
 IPT | 31.61 | 0.8038 | 29.60 | 0.7382 | 37.63 | 0.9580 | 32.45 | 0.9213 |
| Â | Finetuned | |||||||
 MR-IPT | 34.52 | 0.8681 | 31.45 | 0.7952 | 42.48 | 0.9831 | 35.53 | 0.9557 |
 IPT | 32.70 | 0.8094 | 30.03 | 0.7469 | 39.43 | 0.9640 | 34.85 | 0.9419 |
