表 1. Quantitative evaluation of the effect of attention map on different variants of multi-layer network encoders.
注意力图作用在网络编码器多层次不同变体的定量评价
| 网络结构 | DSC | L-pre | R-pre | L-recall | R-recall |
| U-net | 87.54% | 83.24% | 80.41% | 84.91% | 84.04% |
| AGD(1) | 89.31% | 83.47% | 81.19% | 85.98% | 85.25% |
| AGD(1～2) | 89.36% | 86.99% | 82.13% | 82.77% | 85.13% |
| AGD(1～3) | 89.46% | 85.59% | 82.35% | 83.88% | 84.63% |
| AGD(1～4) | 89.60% | 87.34% | 85.75% | 84.15% | 82.63% |
| AGD(1～5, X3 + X4) | 88.51% | 83.18% | 79.33% | 82.94% | 83.62% |
| AGD(1～5, X4 + X5) | 89.43% | 85.64% | 81.72% | 84.39% | 84.58% |
| AGD(1～5) (本文) | 89.97% | 87.17% | 82.84% | 86.54% | 86.48% |