Chest X-ray classification on CheXpert (test)
89.8AUROC (Macro)ViT-B-LR
Evaluation results
| Method | Links | |||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ViT-B-LRArchitecture=ViT-B/16, Rank=0.05r2024.02 | 89.8 | — | — | — | — | — | — | — | — | 85.4 | 94.6 | — | — | — | — | — | — | — | — | |
| ViT-S-LRArchitecture=ViT-S/16, Rank=0.05r2024.02 | 89.6 | — | — | — | — | — | — | — | — | 86.3 | 93.7 | — | — | — | — | — | — | — | — | |
| Pham et al.Architecture=null, Rank=null2024.02 | 89.4 | — | — | — | — | — | — | — | — | 82.5 | 85.5 | — | — | — | — | — | — | — | — | |
| ViT-BArchitecture=ViT-B/16, Rank=null2024.02 | 89.3 | — | — | — | — | — | — | — | — | 82.7 | 83.5 | — | — | — | — | — | — | — | — | |
| ViT-SArchitecture=ViT-S/16, Rank=null2024.02 | 89.2 | — | — | — | — | — | — | — | — | 83.5 | 81.8 | — | — | — | — | — | — | — | — | |
| Kang et al.Architecture=DN121, Rank=null2024.02 | 89 | — | — | — | — | — | — | — | — | 82.1 | 85.9 | — | — | — | — | — | — | — | — | |
| Irvin et al.Architecture=null, Rank=null2024.02 | 88.9 | — | — | — | — | — | — | — | — | 81.8 | 82.8 | — | — | — | — | — | — | — | — | |
| DN121 (MoCo v2)Architecture=DN121, Rank=null2024.02 | 88.7 | — | — | — | — | — | — | — | — | 78.5 | 77.9 | — | — | — | — | — | — | — | — | |
| DN121Architecture=DN121, Rank=null2024.02 | 88.7 | — | — | — | — | — | — | — | — | 81.5 | 77.6 | — | — | — | — | — | — | — | — | |
| ResNet44 + SESE module=true, Downsampling factor f=82025.06 | 88.5 | 92.1 | 89.5 | 80.9 | 70.4 | 79.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet56 + SESE module=true, Downsampling factor f=82025.06 | 88.5 | 92.4 | 89.6 | 81.2 | 70.8 | 80.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet44SE module=false, Downsampling factor f=82025.06 | 88.4 | 92 | 89.3 | 80.4 | 70.1 | 79.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet32SE module=false, Downsampling factor f=82025.06 | 88.2 | 91.9 | 89.2 | 80.4 | 69.7 | 79.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet32 + SESE module=true, Downsampling factor f=82025.06 | 88.2 | 92 | 89.3 | 80.8 | 70 | 79.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet56SE module=false, Downsampling factor f=82025.06 | 88.2 | 92 | 89.3 | 80.9 | 70 | 79.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet20 + SESE module=true, Downsampling factor f=82025.06 | 88.1 | 92 | 89 | 80.3 | 69.9 | 79.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet20SE module=false, Downsampling factor f=82025.06 | 88 | 91.8 | 88.7 | 80.1 | 69.6 | 79.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Haghighi et al.Architecture=DN121, Rank=null2024.02 | 87.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Seyyedkalantari et al.Architecture=null, Rank=null2024.02 | 87.3 | — | — | — | — | — | — | — | — | 81.2 | 83 | — | — | — | — | — | — | — | — | |
| Hosseinzadeh et al.Architecture=DN121, Rank=null2024.02 | 87.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HefNet + SESE module=true, Downsampling factor f=82025.06 | 87 | 91.1 | 88.3 | 79.8 | 69.4 | 79 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HefNetSE module=false, Downsampling factor f=82025.06 | 86.9 | 90.9 | 87.9 | 79.6 | 69.5 | 78.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VGGNet + SESE module=true, Downsampling factor f=82025.06 | 86.2 | 90.5 | 87.5 | 78.8 | 67.7 | 77.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LeNet + SESE module=true, Downsampling factor f=82025.06 | 85.9 | 90.3 | 87 | 78.4 | 67.8 | 77.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VGGNetSE module=false, Downsampling factor f=82025.06 | 85.6 | 90.1 | 86.8 | 78.1 | 64.8 | 75.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LeNetSE module=false, Downsampling factor f=82025.06 | 83.5 | 89.2 | 85 | 76.6 | 60.6 | 74.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Allaouzi et al.Architecture=null, Rank=null2024.02 | 82.8 | — | — | — | — | — | — | — | — | 72 | 88 | — | — | — | — | — | — | — | — | |
| Allaouzi et al. BRUtilization of Segmentation Masks=false2021.06 | — | — | — | — | — | — | — | — | — | 72 | 88 | 87 | 77 | 90 | 82.8 | — | — | — | — | |
| Anatomy-XNetUtilization of Segmentation Masks=true, Input Resolution=224x2242021.06 | — | — | — | — | — | — | — | — | — | 86.55 | 87.86 | 95.28 | 93.13 | 94.66 | 91.5 | — | — | — | — | |
| Anatomy-XNetUtilization of Segmentation Masks=true, Input Resolution=512x5122021.06 | — | — | — | — | — | — | — | — | — | 86.72 | 89.54 | 95.73 | 93.31 | 95.04 | 92.07 | — | — | — | — | |
| Arias-Garzón et al.Utilization of Segmentation Masks=true2021.06 | — | — | — | — | — | — | — | — | — | 81.74 | 84.24 | 94.06 | 90.74 | 94.31 | 89.02 | — | — | — | — | |
| Baseline2025.03 | — | — | — | — | — | — | 80 | — | — | 81 | 86 | — | 73 | 87 | — | 74 | — | — | — | |
| CheXzeroCLIP pretraining data=MIMIC, setting=zero-shot2023.03 | — | — | — | — | — | — | 74.73 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ConvNeXt-BInitialization strategy=ImageNet, Resolution=224x224, Evaluation protocol=full fine-tuning2025.10 | — | — | — | — | — | — | 79.5 | — | — | — | — | — | — | — | — | — | — | — | 0.001 | |
| ConvNeXt-BInitialization strategy=DINOv3, Resolution=224x224, Evaluation protocol=full fine-tuning2025.10 | — | — | — | — | — | — | 80.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ConvNeXt-BInitialization strategy=ImageNet, Resolution=512x512, Evaluation protocol=full fine-tuning2025.10 | — | — | — | — | — | — | 81.6 | — | — | — | — | — | — | — | — | — | — | — | 0.001 | |
| ConvNeXt-BInitialization strategy=DINOv3, Resolution=512x512, Evaluation protocol=full fine-tuning2025.10 | — | — | — | — | — | — | 82.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DAMensemble=5 models, loss=AUC-M loss, leaderboard_name=DeepAUC-v12020.12 | — | — | — | — | — | — | 93.05 | 2.8 | 1 | — | — | — | — | — | — | — | — | — | — | |
| DAMloss=AUC square loss2020.12 | — | — | — | — | — | — | 92.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepAUCParams=6M, view_mode=single-view2022.04 | — | — | — | — | — | — | 83.4 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepAUCParams=6M, view_mode=multi-view2022.04 | — | — | — | — | — | — | 82 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Finetune2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 33 | — | 35.7 | 9.6 | — | |
| Hierarchical Learningtraining loss=CE loss, technique=domain knowledge disease hierarchy2020.12 | — | — | — | — | — | — | 92.99 | 2.6 | 2 | — | — | — | — | — | — | — | — | — | — | |
| HyDA2025.03 | — | — | — | — | — | — | 82 | — | — | 82 | 85 | — | 82 | 89 | — | 74 | — | — | — | |
| Irvin et al. U-OnesUtilization of Segmentation Masks=false2021.06 | — | — | — | — | — | — | — | — | — | 85.8 | 83.2 | 94.1 | 89.9 | 93.4 | 89.3 | — | — | — | — | |
| Keidar et al.Utilization of Segmentation Masks=true2021.06 | — | — | — | — | — | — | — | — | — | 86.42 | 87.39 | 91.97 | 88.23 | 91.73 | 89.15 | — | — | — | — | |
| Lipschitz2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 38 | — | 37.6 | 0 | — | |
| MANetUtilization of Segmentation Masks=true2021.06 | — | — | — | — | — | — | — | — | — | 81.35 | 86.61 | 92.22 | 91.59 | 89.86 | 88.33 | — | — | — | — | |
| MDAN2025.03 | — | — | — | — | — | — | 76 | — | — | 77 | 76 | — | 71 | 84 | — | 72 | — | — | — | |
| Nabla Tau2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 27 | — | 28.5 | 4.4 | — | |
| Original (SF(DA)2)2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 32 | — | 36.9 | 16 | — | |
| PADA2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 36 | — | 36.1 | 0 | — | |
| Pham et al. U-Ones+CT+LSRUtilization of Segmentation Masks=false2021.06 | — | — | — | — | — | — | — | — | — | 82.5 | 85.5 | 93 | 93.7 | 92.3 | 89.4 | — | — | — | — | |
| PHDeepAUCParams=3M, n=22022.04 | — | — | — | — | — | — | 86.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| PHResNet18Params=5M2022.04 | — | — | — | — | — | — | 72.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet18Params=11M2022.04 | — | — | — | — | — | — | 64.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Retrain2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 40 | — | 39.6 | 0 | — | |
| SCADA-UL (Ours)2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 38 | — | 38.1 | 0 | — | |
| SHOT2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 26 | — | 31.1 | 20.2 | — | |
| Stanford Baseline2020.12 | — | — | — | — | — | — | 90.65 | 1.8 | 85 | — | — | — | — | — | — | — | — | — | — | |
| TENT2025.03 | — | — | — | — | — | — | 81 | — | — | 76 | 86 | — | 77 | 89 | — | 76 | — | — | — | |
| Unlearned(+)2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 36 | — | 42 | 15.1 | — | |
| UNSIR2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 27 | — | 27.5 | 0 | — | |
| ViT-BInitialization strategy=ImageNet, Resolution=224x224, Evaluation protocol=full fine-tuning2025.10 | — | — | — | — | — | — | 79.7 | — | — | — | — | — | — | — | — | — | — | — | 0.006 | |
| ViT-BInitialization strategy=DINOv2, Resolution=224x224, Evaluation protocol=full fine-tuning2025.10 | — | — | — | — | — | — | 80.3 | — | — | — | — | — | — | — | — | — | — | — | 0.006 | |
| ViT-BInitialization strategy=DINOv3, Resolution=224x224, Evaluation protocol=full fine-tuning2025.10 | — | — | — | — | — | — | 80 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-BInitialization strategy=ImageNet, Resolution=512x512, Evaluation protocol=full fine-tuning2025.10 | — | — | — | — | — | — | 80.7 | — | — | — | — | — | — | — | — | — | — | — | 0.006 | |
| ViT-BInitialization strategy=DINOv2, Resolution=512x512, Evaluation protocol=full fine-tuning2025.10 | — | — | — | — | — | — | 81.4 | — | — | — | — | — | — | — | — | — | — | — | 0.006 | |
| ViT-BInitialization strategy=DINOv3, Resolution=512x512, Evaluation protocol=full fine-tuning2025.10 | — | — | — | — | — | — | 81.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| XplainerCLIP pretraining data=MIMIC, setting=zero-shot2023.03 | — | — | — | — | — | — | 80.58 | — | — | — | — | — | — | — | — | — | — | — | — | |
| YWWpooling operator=Probabilistic-CAM (PCAM), training loss=CE loss, technique=weakly-supervised lesion localization2020.12 | — | — | — | — | — | — | 92.89 | 2.8 | 5 | — | — | — | — | — | — | — | — | — | — | |
| ZSMU2026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 27 | — | 26.9 | 0 | — |