RT Dose Prediction on GDP-HMM (val)
1.89MAEDiffKT3D
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DiffKT3DPretraining=MAISI, CT, Sampling steps=10-step2026.05 | 1.89 | 135.89 | — | — | — | |
| MAISI + OursBackbone=LDM, Pretrain=true, Any2Any=false, RL=false2026.05 | 1.89 | 135.89 | 32.02 | 97.5 | 0.028 | |
| DiffKT3DParadigm=Any2Any, Fine-tuning=full fine-tuning, Sampling steps=10-step2026.05 | 1.9 | 136.22 | — | — | — | |
| Ours (Any2Any)Backbone=DiT, Pretrain=true, Any2Any=true, RL=false2026.05 | 1.9 | 136.22 | 32.43 | 97.7 | 0.025 | |
| DiffKT3D + ScardNFTParadigm=Any2Any, Post-training=ScardNFT, Sampling steps=10-step2026.05 | 1.91 | 138.17 | — | — | — | |
| Ours (Any2Any+NFT)Backbone=DiT, Pretrain=true, Any2Any=true, RL=true2026.05 | 1.91 | 138.17 | 32.55 | 97.9 | 0.022 | |
| Challenge Top-1Method Type=regression2026.05 | 2.03 | 134.26 | — | — | — | |
| Yasin (AAPM ’25 Chal.)Backbone=MedNeXt, Pretrain=false, Any2Any=false, RL=false2026.05 | 2.03 | 134.26 | 31.5 | 97.2 | 0.037 | |
| rcgao (AAPM ’25 Chal.)Backbone=MedNeXt, Pretrain=false, Any2Any=false, RL=false2026.05 | 2.05 | 133.4 | 31.24 | 97.2 | 0.051 | |
| DiffKT3DArchitecture=Conditional DiT, Sampling steps=10-step2026.05 | 2.07 | 135.41 | — | — | — | |
| Ours (Conditional)Backbone=DiT, Pretrain=true, Any2Any=false, RL=false2026.05 | 2.07 | 135.41 | 31.88 | 97.3 | 0.031 | |
| 2D slice-wise diffusionSampling steps=10-step2026.05 | 2.14 | 132.9 | — | — | — | |
| PVmed (AAPM ’25 Chal.)Backbone=MedNeXt, Pretrain=false, Any2Any=false, RL=false2026.05 | 2.15 | 133.78 | 31.4 | 97.1 | 0.034 | |
| tyxiong123 (AAPM ’25 Chal.)Backbone=MedNeXt, Pretrain=false, Any2Any=false, RL=false2026.05 | 2.18 | 133.95 | 29.89 | 92.9 | 0.067 | |
| MedVision (AAPM ’25 Chal.)Backbone=LDM, Pretrain=false, Any2Any=false, RL=false2026.05 | 2.19 | 134.5 | 31.54 | 97 | 0.029 | |
| SKLSDE-BH (AAPM ’25 Chal.)Backbone=nnUnet, Pretrain=false, Any2Any=false, RL=false2026.05 | 2.25 | 133.02 | 31.4 | 97.2 | 0.04 | |
| LoRA on Wan DiT + Any2AnyBackbone=Wan 2.1 DiT, Fine-tuning=rank-64 LoRA adapters, Sampling steps=10-step2026.05 | 2.26 | 132.44 | — | — | — | |
| 3D ControlNetBackbone=Wan 2.1 DiT, Sampling steps=10-step2026.05 | 2.42 | 125.79 | — | — | — |