Disease Classification on CheXpert
0.9238AUROCCARZero
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CARZeroZero-shot=true2026.05 | 0.9238 | 40.2 | 86.93 | |
| KEPILZero-shot=true2026.05 | 0.9121 | 65.56 | 87.88 | |
| X-WINPretrain. data=372K CXRs + 32K CT volumes, Backbone=ViT-Large, Evaluation protocol=linear probing2025.11 | 0.908 | — | — | |
| X-WINPretrain. data=372K CXRs + 32K CT volumes, Backbone=ViT-Base, Evaluation protocol=linear probing2025.11 | 0.902 | — | — | |
| MAVLZero-shot=true2026.05 | 0.9013 | 65.47 | 86.44 | |
| DeViDeZero-shot=true2026.05 | 0.8987 | 62.88 | 87.98 | |
| MGCADirect Inference=Yes, Labeling ratio=100%2026.06 | 0.897 | — | — | |
| KADZero-shot=true2026.05 | 0.8923 | 63.25 | 86.25 | |
| Medical MAEDirect Inference=No, Labeling ratio=100%2026.06 | 0.892 | — | — | |
| IMT-CXRDirect Inference=Yes, Labeling ratio=100%2026.06 | 0.892 | — | — | |
| MGCADirect Inference=Yes, Labeling ratio=10%2026.06 | 0.891 | — | — | |
| IMT-CXRDirect Inference=Yes, Labeling ratio=10%2026.06 | 0.89 | — | — | |
| MGCADirect Inference=Yes, Labeling ratio=1%2026.06 | 0.888 | — | — | |
| MRMDirect Inference=No, Labeling ratio=100%2026.06 | 0.887 | — | — | |
| MRMDirect Inference=No, Labeling ratio=1%2026.06 | 0.885 | — | — | |
| MRMDirect Inference=No, Labeling ratio=10%2026.06 | 0.885 | — | — | |
| IMT-CXRDirect Inference=Yes, Labeling ratio=1%2026.06 | 0.885 | — | — | |
| REFERSDirect Inference=No, Labeling ratio=100%2026.06 | 0.882 | — | — | |
| REFERSDirect Inference=No, Labeling ratio=10%2026.06 | 0.881 | — | — | |
| GloRIADirect Inference=Yes, Labeling ratio=100%2026.06 | 0.881 | — | — | |
| MedKLIPZero-shot=true2026.05 | 0.8797 | 63.67 | 84.32 | |
| CheXzeroZero-shot=true2026.05 | 0.879 | 61.9 | 81.17 | |
| M3AEDirect Inference=No, Labeling ratio=100%2026.06 | 0.879 | — | — | |
| GloRIADirect Inference=Yes, Labeling ratio=10%2026.06 | 0.878 | — | — | |
| CheXFoundPretrain. data=987K CXRs, Backbone=ViT-Large, Evaluation protocol=linear probing2025.11 | 0.876 | — | — | |
| Ark+Pretrain. data=704K CXRs, Backbone=Swin-Base, Evaluation protocol=linear probing2025.11 | 0.876 | — | — | |
| M3AEDirect Inference=No, Labeling ratio=10%2026.06 | 0.873 | — | — | |
| ConVIRTDirect Inference=Yes, Labeling ratio=100%2026.06 | 0.873 | — | — | |
| REFERSDirect Inference=No, Labeling ratio=1%2026.06 | 0.872 | — | — | |
| CheXWorldPretrain. data=448K CXRs, Backbone=ViT-Base, Evaluation protocol=linear probing2025.11 | 0.871 | — | — | |
| ConVIRTDirect Inference=Yes, Labeling ratio=10%2026.06 | 0.868 | — | — | |
| GloRIADirect Inference=Yes, Labeling ratio=1%2026.06 | 0.866 | — | — | |
| MaCoPretrain. data=377K CXRs w/ reports, Backbone=ViT-Base, Evaluation protocol=linear probing2025.11 | 0.864 | — | — | |
| M3AEDirect Inference=No, Labeling ratio=1%2026.06 | 0.862 | — | — | |
| ConVIRTDirect Inference=Yes, Labeling ratio=1%2026.06 | 0.859 | — | — | |
| CARZeroPretrain. data=377K CXRs w/ reports, Backbone=ViT-Base, Evaluation protocol=linear probing2025.11 | 0.852 | — | — | |
| RAD-DINOPretrain. data=LVD-142M + 838K CXRs, Backbone=ViT-Base, Evaluation protocol=linear probing2025.11 | 0.846 | — | — | |
| CheXAgentPretrain. data=1.07M CXRs w/ text, Backbone=ViT-Large, Evaluation protocol=linear probing2025.11 | 0.813 | — | — | |
| CXR-AlignPretrain. data=325K CXRs w/ reports, Backbone=ViT-Base, Evaluation protocol=linear probing2025.11 | 0.793 | — | — | |
| DINOv2Pretrain. data=LVD-142M, Backbone=ViT-Base, Evaluation protocol=linear probing2025.11 | 0.776 | — | — | |
| I-JEPAPretrain. data=ImageNet-1K, Backbone=ViT-Large, Evaluation protocol=linear probing2025.11 | 0.764 | — | — | |
| BioViL-TZero-shot=true2026.05 | 0.7093 | 47.21 | 69.96 | |
| BioViLZero-shot=true2026.05 | 0.6001 | 42.1 | 66.13 | |
| GLoRIAZero-shot=true2026.05 | 0.5484 | 37.86 | 60.7 | |
| ConVIRTZero-shot=true2026.05 | 0.521 | 35.61 | 57.43 |