Table 1.
Quantitative comparisons of the low-count PET denoised images between our FedFTN and locally trained models. The local single model means FTN-modulated denoising networks trained at one specific low-count level at the specific institution. The local unified model means FTN-modulated denoising networks trained with all three low-count levels within each institution. The performance of FedFTN with Site Adaptation (SA) via further local data fine-tuning is reported in the last row. The best results are marked in bold.
| Evaluation PSNR/NMSE/SSIM | Institution #1 | Institution #2 | Institution #3 | ||||||
|---|---|---|---|---|---|---|---|---|---|
|
| |||||||||
| 5% | 10% | 20% | 2% | 5% | 10% | 2% | 5% | 10% | |
|
| |||||||||
| Original | 20.46/.129/.931 | 23.80/.059/.954 | 27.40/.026/.972 | 19.90/.076/.923 | 23.66/.031/.951 | 26.26/.017/.968 | 20.61/.121/.933 | 24.89/.050/.961 | 27.55/.030/.975 |
| Local Single Models | 25.93/.034/.966 | 27.91/.021/.975 | 30.22/.012/.983 | 25.37/.021/.971 | 27.16/.014/.980 | 28.64/.010/.985 | 26.00/.035/.974 | 28.17/.023/.982 | 29.90/.018/.986 |
| Local Unified Model | 26.02/.032/.968 | 28.01/.020/.975 | 30.21/.012/.983 | 25.42/.020/.970 | 27.21/.014/.980 | 28.64/.010/.985 | 26.02/.035/.975 | 28.20/.023/.982 | 29.91/.018/.986 |
|
| |||||||||
| FedFTN | 27.24/.025/.999 † | 28.96/.017/.999 † | 30.82/.011/.999 † | 26.12/.018/.999 † | 27.80/.013/.999 † | 29.03/.009/.999 † | 26.83/.031/.999 † | 28.91/.021/.999 † | 30.23/.017/.999 † |
”†”
indicates that the difference between FedFTN and all compared methods is significant at p < 0.005 based on the non-parametricWilcoxon signed rank test.