Medical Image Segmentation on MSD Heart (5-fold CV)
93.28Dice ScoreCPS
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| CPSScheme=Constant Patch Size Training Scheme, Evaluation Protocol=5-fold Cross-Validation2024.07 | 93.28 | — | — | 15.04 | 100 | 6.24 | |
| PGPS+Scheme=PGPS with increased batch size, Evaluation Protocol=5-fold Cross-Validation2024.07 | 93.28 | — | 0.4763 | 7.06 | 39.35 | 2.89 | |
| PGPSScheme=Progressive Growing of Patch Size Training Scheme, Evaluation Protocol=5-fold Cross-Validation2024.07 | 93.21 | — | 0.2567 | 6.44 | 35.35 | 2.64 | |
| AdamCPROptimizer=AdamCPR, Warm start steps (s)=2k, Learning rate warmup=2k, Total training steps=25k, Initialization=Kappa-WS2023.11 | 93.18 | — | — | — | — | — | |
| AdamCPROptimizer=AdamCPR, Warm start steps (s)=3k, Learning rate warmup=2k, Total training steps=25k, Initialization=Kappa-WS2023.11 | 93.16 | — | — | — | — | — | |
| AdamWOptimizer=AdamW, Weight decay (γ)=1e-5, Learning rate warmup=2k, Total training steps=25k2023.11 | 92.92 | — | — | — | — | — | |
| AdamWOptimizer=AdamW, Weight decay (γ)=1e-2, Learning rate warmup=2k, Total training steps=25k2023.11 | 92.9 | — | — | — | — | — | |
| AdamWOptimizer=AdamW, Weight decay (γ)=1e-3, Learning rate warmup=2k, Total training steps=25k2023.11 | 92.88 | — | — | — | — | — | |
| AdamWOptimizer=AdamW, Weight decay (γ)=1e-1, Learning rate warmup=2k, Total training steps=25k2023.11 | 92.85 | — | — | — | — | — | |
| AdamCPROptimizer=AdamCPR, Warm start steps (s)=1k, Learning rate warmup=2k, Total training steps=25k, Initialization=Kappa-WS2023.11 | 92.77 | — | — | — | — | — | |
| AdamWOptimizer=AdamW, Weight decay (γ)=1e-4, Learning rate warmup=2k, Total training steps=25k2023.11 | 92.75 | — | — | — | — | — | |
| AdamCPROptimizer=AdamCPR, Warm start steps (s)=4k, Learning rate warmup=2k, Total training steps=25k, Initialization=Kappa-WS2023.11 | 74.44 | — | — | — | — | — | |
| 2D U-NetArchitecture=2D U-Net2018.09 | — | 91.36 | — | — | — | — | |
| 3D U-NetArchitecture=3D U-Net2018.09 | — | 92.45 | — | — | — | — | |
| 3D U-Net (U-Net Cascade)Architecture=3D U-Net, Pipeline=U-Net Cascade2018.09 | — | 92.4 | — | — | — | — | |
| 3D U-Net stage1 onlyArchitecture=3D U-Net, Pipeline=U-Net Cascade, Stage=1 only2018.09 | — | 90.63 | — | — | — | — | |
| ensemble 2D U-Net + 3D U-NetEnsemble=2D U-Net + 3D U-Net2018.09 | — | 92.7 | — | — | — | — | |
| ensemble 2D U-Net + 3D U-Net (U-Net Cascade)Ensemble=2D U-Net + 3D U-Net (Cascade)2018.09 | — | 92.64 | — | — | — | — | |
| ensemble 3D U-Net + 3D U-Net (U-Net Cascade)Ensemble=3D U-Net + 3D U-Net (Cascade)2018.09 | — | 92.63 | — | — | — | — |