Table 1 Volume-based evaluation for all models evaluated. There are 98 patients with growing (GRW) and 54 with shrinking (SHR) volume of WMH. “CP” stands for “Correct Prediction”, “CPinEVI” stands for “Correct Prediction in Estimated Volume Interval”, and “(CP+WP)inEVI” stands for “Correct Prediction + Wrong Prediction but still in EVI”. Symbol ↑ indicates higher values are better, while symbol → 0 indicates that values closer to 0 are better. Each evaluation measurement’s best and second-best values are written in bold and underlined, respectively. PUNet-wSL-vol model is highlighted as it emerged as the best-performing model to estimate the future volume of WMH.
Model’s name | Prediction↑ | Estimated volume interval (n = 152) ↑ | Volumetric error | |||
|---|---|---|---|---|---|---|
GRW | SHR | CP | CPinEVI | (CP + WP)inEVI | (std) → 0 | |
UNet43 | 79.59% | 66.67% | 67.11% | 48.03% | 58.55% | 1.267 (8.623) |
UNet-vol | 80.61% | 68.52% | 67.11% | 46.05% | 55.26% | − 0.194 (8.107) |
UNet-wSL | 72.45% | 64.81% | 71.71% | 38.16% | 47.37% | 1.038 (9.427) |
UNet-wSL-vol | 84.69% | 59.26% | 71.71% | 48.68% | 59.87% | 0.027 (8.662) |
PUNet26 | 78.57% | 46.30% | 67.11% | 47.37% | 61.18% | − 1.774 (9.798) |
PUNet-vol | 83.67% | 51.85% | 71.71% | 46.71% | 60.53% | − 0.834 (8.657) |
PUNet-wSL | 75.51% | 64.81% | 71.71% | 48.68% | 59.21% | 0.227 (10.427) |
PUNet-wSL-vol | 74.49% | 74.07% | 74.34% | 53.29% | 62.50% | − 0.009 (9.751) |
Att-PUNet | 70.41% | 79.63% | 73.68% | 45.39% | 55.26% | 3.182 (8.447) |
Att-PUNet-vol | 81.63% | 55.56% | 72.37% | 43.42% | 54.61% | − 0.555 (9.043) |
Att-PUNet-wSL | 86.73% | 55.56% | 75.66% | 51.97% | 59.87% | − 0.598 (10.901) |
Att-PUNet-wSL-vol | 81.63% | 64.81% | 75.66% | 43.42% | 53.95% | 0.270 (9.050) |
