Table 3.
Quantitative analysis of the proposed method (MS-RDNZ5) and the FIRST algorithm.
| Metrics | FDK300 reference | FIRST300 reference | ||
|---|---|---|---|---|
| FIRST100 | MS-RDNZ5 | FIRST100 | MS-RDNZ5 | |
| NMSE (dB) | ||||
| S | 27.19 (1.03) | 27.50 (1.03) | 32.67 (1.01) | 32.99 (1.17) |
| M | 24.85 (0.19) | 25.67 (0.31) | 29.15 (0.36) | 31.45 (1.53) |
| L | 26.11 (0.38) | 26.78 (0.29) | 34.93 (0.52) | 35.68 (0.71) |
| Bias ( cm) | ||||
| S | 9.08 (1.24) | 8.60 (1.65) | 4.55 (0.71) | 4.12 (0.81) |
| M | 11.80 (0.21) | 10.69 (0.29) | 6.98 (0.26) | 5.19 (0.79) |
| L | 8.68 (0.34) | 8.07 (0.19) | 2.92 (0.19) | 2.75 (0.26) |
| PSNR (dB) | ||||
| S | 41.17 (1.17) | 41.42 (1.37) | 46.77 (1.09) | 47.00 (1.28) |
| M | 38.95 (0.16) | 39.80 (0.25) | 43.23 (0.31) | 45.56 (1.43) |
| L | 41.62 (0.32) | 42.28 (0.21) | 50.45 (0.53) | 51.22 (0.73) |
| SSIM | ||||
| S | 0.941 (0.020) | 0.946 (0.021) | 0.988 (0.004) | 0.989 (0.003) |
| M | 0.893 (0.004) | 0.914 (0.003) | 0.964 (0.003) | 0.985 (0.002) |
| L | 0.938 (0.003) | 0.944 (0.002) | 0.994 (0.001) | 0.994 (0.001) |
One small-size breast (S), one medium-size breast (M), and one large-size breast (L) were selected for testing, respectively. The suffixes “100” and “300” denote the number of projections in the data. The MS-RDNZ5 network was always trained using FDK100 as input and FDK300 as label. However, either FDK300 or FIRST300 were used as the reference when computing the quality metrics, as indicated by the column labels “FDK300 Reference” and “FIRST300 Reference”, respectively.
Median and interquartile range in the bracket are shown.
Bolded values indicate better performance in pairwise comparison.