Table 2 Results of different training strategies on the LUNA dataset.
From: Lessons learned from RadiologyNET foundation models for transfer learning in medical radiology
Pretrain strategy | TL model | LR | Dice score | IoU | |
|---|---|---|---|---|---|
U-Net | R | ImageNet | \(10^{-4}\) | 0.111 ± 0.002 | 0.063 ± 0.002 |
R | RadiologyNET | \(10^{-4}\) | 0.111 ± 0.002 | 0.063 ± 0.001 | |
N/A | Baseline | \(10^{-4}\) | 0.616 ± 0.012 | 0.500 ± 0.011 | |
U-Net- EfficientNetB4 | C | ImageNet | \(10^{-4}\) | 0.685 ± 0.016 | 0.582 ± 0.017 |
C | RadiologyNET | \(10^{-4}\) | 0.695 ± 0.022 | 0.593 ± 0.026 | |
N/A | Baseline | \(10^{-4}\) | 0.688 ± 0.013 | 0.586 ± 0.014 | |
U-Net- ResNet50 | C | ImageNet | \(10^{-4}\) | 0.692 ± 0.026 | 0.593 ± 0.03 |
C | RadiologyNET | \(10^{-5}\) | 0.715 ± 0.017 | 0.616 ± 0.017 | |
N/A | Baseline | \(10^{-4}\) | 0.646 ± 0.027 | 0.538 ± 0.03 | |
U-Net- VGG16 | C | ImageNet | \(10^{-5}\) | 0.729 ± 0.01 | 0.632 ± 0.015 |
C | RadiologyNET | \(10^{-4}\) | 0.706 ± 0.015 | 0.605 ± 0.019 | |
N/A | Baseline | \(10^{-4}\) | 0.704 ± 0.03 | 0.601 ± 0.033 |
