Table 1.
Quantitative performance of the DL-ToF models evaluated on 50 test exams, expressed as a percentage difference with ToF BSREM (taken as ground truth), for lesion SUVmax, lung SUVmean, liver SUVmean, and noise in liver (the standard deviation of noise averaged over all exams) for each type of reconstruction. P-values (parentheses) show significance of difference from ToF BSREM
| Lesion SUVmax (%) | Lung SUVmean (%) | Liver SUVmean (%) | Liver noise (SUV) | |
|---|---|---|---|---|
| Non-ToF BSREM | − 28.6 ± 18.3 (< 0.0001) | 7.7 ± 15.0 (< 0.0001) | 4.3 ± 5.6 (< 0.0001) | 0.16 |
| DL-ToF(L) | − 28.7 ± 19.0 (< 0.0001) | 0.6 ± 12.1 (0.179) | 0.7 ± 4.6 (0.067) | 0.10 |
| DL-ToF(M) | − 8.0 ± 22.5 (< 0.0001) | 1.3 ± 13.0 (0.083) | 0.8 ± 4.4 (0.016) | 0.13 |
| DL-ToF(H) | 1.7 ± 23.9 (0.57) | 1.4 ± 11.5 (0.50) | 0.1 ± 4.5 (0.86) | 0.19 |
| ToF-BSREM | – | – | – | 0.19 |