Thoracic Disease Classification on MIMIC-CXR (test)
84.04Average AUCAnatomy-XNet
Evaluation results
| Method | Links | |||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Anatomy-XNetUtilization of Segmentation Masks=true, Input Resolution=512x5122021.06 | 84.04 | 83.93 | 82.59 | 84.84 | 90.76 | 75.12 | 74.95 | 78.78 | 78.9 | 86.97 | 93.43 | 86.21 | 75.81 | 91.2 | 93.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Anatomy-XNetUtilization of Segmentation Masks=true, Input Resolution=224x2242021.06 | 83.9 | 83.79 | 82.67 | 85.25 | 90.83 | 75.45 | 74.3 | 77.08 | 78.79 | 86.9 | 93.37 | 86.55 | 75.98 | 90.87 | 92.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ChexclusionUtilization of Segmentation Masks=false, Ensemble=5 checkpoints2021.06 | 83.4 | 83.7 | 82.8 | 84.4 | 90.4 | 75.7 | 71.8 | 77.2 | 78.2 | 86.8 | 93.3 | 84.8 | 74.8 | 90.3 | 92.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Keidar et al.Utilization of Segmentation Masks=true2021.06 | 82.9 | 83.24 | 82.59 | 84.19 | 90.4 | 74.71 | 71.33 | 76.66 | 77.67 | 86.39 | 92.93 | 84.18 | 74.51 | 89.7 | 92.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MANetUtilization of Segmentation Masks=true2021.06 | 81.92 | 82.77 | 81.86 | 83.66 | 90.03 | 74.52 | 69.56 | 75.43 | 77.24 | 85.9 | 91.53 | 83.05 | 73.01 | 88.02 | 90.24 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Arias-Garzón et al.Utilization of Segmentation Masks=true2021.06 | 81.33 | 82.61 | 81.57 | 83.16 | 90.01 | 73.71 | 65.36 | 74.57 | 77.4 | 85.83 | 91.5 | 81.96 | 72.79 | 87.47 | 90.64 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Global-Local Contrastive Learning (augmented)Training Data=MIMIC-CXR + PadChest + ChestX-Ray14, Prompting Scheme=detailed2022.05 | 80.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| XpertCausalBackbone=InceptionV32026.05 | 79.68 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 71.09 | 0.0079 | |
| Global-Local Contrastive LearningTraining Data=MIMIC-CXR, Prompting Scheme=detailed2022.05 | 79.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LoCoTraining Data=MIMIC-CXR, Prompting Scheme=detailed, Objective=Local2022.05 | 78.15 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Label-SupervisedTraining Data=MIMIC-CXR, Supervision=Labels2022.05 | 77.42 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GloCoTraining Data=MIMIC-CXR, Prompting Scheme=detailed, Objective=Global2022.05 | 76.58 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Causal (Learned)Backbone=InceptionV3, Configuration=learned concept-pathology associations2026.05 | 76.31 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 68.46 | 0.0102 | |
| XpertXAIBackbone=InceptionV3, Type=non-causal concept-based model2026.05 | 73.61 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 68.02 | 0.0249 | |
| SLIPTraining Data=MIMIC-CXR, Prompting Scheme=detailed2022.05 | 72.44 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VSE-GCNUtilization of Segmentation Masks=false2021.06 | 72.1 | 72.2 | 73 | 72.8 | 79.9 | 76.7 | 56 | 62.3 | 65.4 | 81.7 | 86.3 | 65.3 | 58.8 | 79.7 | 78.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MedB. (MoE)Protocol=Zero-shot cross-dataset, Module=MoE2025.05 | 71.92 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MedProbCLIPmode=zero-shot, prompt=a chest X-ray with {class}2026.02 | 71.01 | 56.5 | 79.47 | 72.92 | 79.86 | 71.63 | 44.55 | 79.76 | 57.67 | — | 86.61 | 72.73 | 73.08 | 78.02 | 70.28 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MedBridgeProtocol=Zero-shot cross-dataset2025.05 | 70.59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPTraining Data=MIMIC-CXR, Prompting Scheme=detailed2022.05 | 70.25 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MILNCElocalTraining Data=MIMIC-CXR, Prompting Scheme=detailed2022.05 | 69.18 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Densenet-KGUtilization of Segmentation Masks=false2021.06 | 68.5 | 69.4 | 74.6 | 64 | 79 | 65.1 | 60.5 | 57.4 | 60.9 | 77.8 | 80.9 | 65 | 57.2 | 68.9 | 78.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InceptionV3Type=standard end-to-end baseline2026.05 | 68.33 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 61.2 | 0.0377 | |
| CXR-CLIPmode=zero-shot, prompt=a chest X-ray with {class}2026.02 | 66.19 | 50 | 79.08 | 47.22 | 81.64 | 78.37 | 48.18 | 72.62 | 81.67 | — | 59.26 | 45.45 | 71.44 | 73.36 | 72.17 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LPProtocol=Zero-shot cross-dataset2025.05 | 65.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPmode=zero-shot, prompt=a chest X-ray with {class}2026.02 | 65.19 | 31 | 78.32 | 47.92 | 84.53 | 66.83 | 63.64 | 48.81 | 75.5 | — | 68.65 | 63.64 | 72.36 | 67.07 | 79.25 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MMRLProtocol=Zero-shot cross-dataset2025.05 | 64.74 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MMAProtocol=Zero-shot cross-dataset2025.05 | 64.03 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CoOpProtocol=Zero-shot cross-dataset2025.05 | 63.69 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| KgCoOpProtocol=Zero-shot cross-dataset2025.05 | 60.55 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CoCoOpProtocol=Zero-shot cross-dataset2025.05 | 59.86 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProGradProtocol=Zero-shot cross-dataset2025.05 | 58.91 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIP-Ad.Protocol=Zero-shot cross-dataset2025.05 | 57.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PCME++mode=zero-shot, prompt=a chest X-ray with {class}2026.02 | 54.56 | 71.5 | 80.61 | 59.72 | 71.26 | 68.75 | 52.73 | 16.67 | 47.5 | — | 50.61 | 45.45 | 46.51 | 46.03 | 51.89 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ALBERT-baseTraining Mode=non-FT (off the shelf model)2025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 51.5 | — | — | |
| ALBERT-baseTraining Mode=DP-LoRA, Privacy Budget (epsilon)=0.01, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 53.9 | — | — | |
| ALBERT-baseTraining Mode=DP-LoRA, Privacy Budget (epsilon)=0.1, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 83.4 | — | — | |
| ALBERT-baseTraining Mode=DP-LoRA, Privacy Budget (epsilon)=1, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 88.6 | — | — | |
| ALBERT-baseTraining Mode=DP-LoRA, Privacy Budget (epsilon)=10, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 89 | — | — | |
| ALBERT-baseTraining Mode=Non-priv LoRA, r=64, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 89.2 | — | — | |
| ALBERT-baseTraining Mode=Non-priv Full FT, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 96.5 | — | — | |
| BERT-mediumTraining Mode=non-FT (off the shelf model)2025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 52.7 | — | — | |
| BERT-mediumTraining Mode=DP-LoRA, Privacy Budget (epsilon)=0.01, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 39.5 | — | — | |
| BERT-mediumTraining Mode=DP-LoRA, Privacy Budget (epsilon)=0.1, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 47 | — | — | |
| BERT-mediumTraining Mode=DP-LoRA, Privacy Budget (epsilon)=1, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 88.3 | — | — | |
| BERT-mediumTraining Mode=DP-LoRA, Privacy Budget (epsilon)=10, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 90.1 | — | — | |
| BERT-mediumTraining Mode=Non-priv LoRA, r=64, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 91 | — | — | |
| BERT-mediumTraining Mode=Non-priv Full FT, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 96.8 | — | — | |
| BERT-smallTraining Mode=non-FT (off the shelf model)2025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 52.3 | — | — | |
| BERT-smallTraining Mode=DP-LoRA, Privacy Budget (epsilon)=0.01, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 37.2 | — | — | |
| BERT-smallTraining Mode=DP-LoRA, Privacy Budget (epsilon)=0.1, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 48 | — | — | |
| BERT-smallTraining Mode=DP-LoRA, Privacy Budget (epsilon)=1, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 89 | — | — | |
| BERT-smallTraining Mode=DP-LoRA, Privacy Budget (epsilon)=10, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 89.8 | — | — | |
| BERT-smallTraining Mode=Non-priv LoRA, r=64, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 90.1 | — | — | |
| BERT-smallTraining Mode=Non-priv Full FT, Epochs=102025.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 96.7 | — | — | |
| CheXNetEvaluation Protocol=images2021.07 | — | 56.7 | 53.4 | 19.3 | 67.4 | — | — | — | — | — | 71.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Class BiasEvaluation Protocol=Statistical baseline2021.07 | — | 24.5 | 22.1 | 6.6 | 17.7 | — | — | — | — | — | 26.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CoDi_XRInput Condition=T→F, Evaluation Model=XRV-DenseNet2025.01 | — | 91 | 95 | 93 | 96 | 84 | — | 82 | 91 | — | 97 | — | 84 | 88 | — | 90 | 90 | 91 | 60 | 60 | 33 | 66 | 10 | 5 | 64 | 78 | 38 | 17 | 53 | 43 | 59 | — | — | |
| CoDi_XRInput Condition=L→F, Evaluation Model=XRV-DenseNet2025.01 | — | 84 | 88 | 90 | 91 | 81 | — | 76 | 85 | — | 95 | — | 76 | 82 | — | 83 | 85 | 90 | 51 | 45 | 32 | 42 | 6 | 7 | 59 | 73 | 27 | 10 | 43 | 35 | 49 | — | — | |
| CoDi_XRInput Condition=T+L→F, Evaluation Model=XRV-DenseNet2025.01 | — | 89 | 91 | 92 | 91 | 82 | — | 80 | 86 | — | 97 | — | 77 | 85 | — | 88 | 89 | 90 | 54 | 55 | 32 | 64 | 9 | 6 | 63 | 75 | 38 | 17 | 53 | 44 | 58 | — | — | |
| LLM-CXRInput Condition=T→F, Evaluation Model=XRV-DenseNet2025.01 | — | 89 | 92 | 87 | 96 | 80 | — | 78 | 85 | — | 95 | — | 80 | 85 | — | 89 | 87 | 87 | 51 | 53 | 19 | 60 | 8 | 7 | 60 | 74 | 38 | 16 | 47 | 39 | 51 | — | — | |
| RATCHETEvaluation Protocol=NLP (generated report classification)2021.07 | — | 41.1 | 44.6 | 4.1 | 40.7 | — | — | — | — | — | 63.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Real DataInput Condition=Original Images, Evaluation Model=XRV-DenseNet2025.01 | — | 84 | 91 | 91 | 93 | 81 | — | 78 | 85 | — | 95 | — | 78 | 86 | — | 87 | 86 | 87 | 49 | 49 | 35 | 60 | 9 | 12 | 60 | 73 | 35 | 21 | 49 | 40 | 53 | — | — | |
| RoentGenInput Condition=T→F, Evaluation Model=XRV-DenseNet2025.01 | — | 87 | 93 | 82 | 80 | 50 | — | 59 | 68 | — | 94 | — | 48 | 58 | — | 78 | 72 | 78 | 31 | 40 | 0 | 25 | 0 | 2 | 40 | 74 | 4 | 4 | 36 | 25 | 43 | — | — | |
| TieNetEvaluation Protocol=Classification2021.07 | — | 47.1 | 48.8 | 18.8 | 53.6 | — | — | — | — | — | 59.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TieNetEvaluation Protocol=NLP (generated report classification)2021.07 | — | 0.6 | 6.1 | 0 | 5 | — | — | — | — | — | 3.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UniXGenInput Condition=T→F, Evaluation Model=XRV-DenseNet2025.01 | — | 75 | 78 | 69 | 81 | 67 | — | 67 | 69 | — | 76 | — | 66 | 66 | — | 74 | 71 | 71 | 25 | 31 | 5 | 29 | 6 | 5 | 37 | 41 | 0 | 4 | 24 | 17 | 34 | — | — | |
| UniXGenInput Condition=T+L→F, Evaluation Model=XRV-DenseNet2025.01 | — | 75 | 79 | 65 | 81 | 66 | — | 63 | 68 | — | 77 | — | 63 | 70 | — | 73 | 71 | 71 | 20 | 31 | 6 | 27 | 6 | 7 | 34 | 43 | 4 | 7 | 24 | 19 | 33 | — | — | |
| XGeMInput Condition=T→F, Evaluation Model=XRV-DenseNet2025.01 | — | 86 | 95 | 91 | 94 | 82 | — | 75 | 86 | — | 97 | — | 86 | 85 | — | 88 | 89 | 90 | 51 | 60 | 25 | 59 | 8 | 11 | 60 | 78 | 33 | 12 | 50 | 41 | 54 | — | — | |
| XGeMInput Condition=L→F, Evaluation Model=XRV-DenseNet2025.01 | — | 86 | 92 | 93 | 94 | 83 | — | 80 | 89 | — | 97 | — | 80 | 90 | — | 87 | 88 | 90 | 51 | 46 | 39 | 57 | 9 | 13 | 64 | 79 | 36 | 17 | 50 | 41 | 54 | — | — | |
| XGeMInput Condition=T+L→F, Evaluation Model=XRV-DenseNet2025.01 | — | 92 | 96 | 95 | 97 | 87 | — | 85 | 92 | — | 98 | — | 85 | 93 | — | 91 | 92 | 92 | 58 | 59 | 39 | 68 | 13 | 18 | 68 | 81 | 40 | 26 | 57 | 47 | 60 | — | — |