Image Classification on CheXpert (test)
89.5AUCEGMA
Evaluation Results
| Method | Links | |
|---|---|---|
| EGMAPortion of training data=100%, Evaluation protocol=Linear classification2024.03 | 89.5 | |
| MGCAPortion of training data=100%, Evaluation protocol=Linear classification2024.03 | 89.3 | |
| PRIORPortion of training data=100%, Evaluation protocol=Linear classification2024.03 | 89.08 | |
| EGMAPortion of training data=10%, Evaluation protocol=Linear classification2024.03 | 88.92 | |
| Centralized (Baseline)Privacy Guarantee=None, Backbone=DenseNet1212025.12 | 88.7 | |
| BioViLPortion of training data=100%, Evaluation protocol=Linear classification2024.03 | 88.62 | |
| GLORIAPortion of training data=100%, Evaluation protocol=Linear classification2024.03 | 88.1 | |
| MedCLIPPortion of training data=100%, Evaluation protocol=Linear classification2024.03 | 88.02 | |
| GLORIAPortion of training data=10%, Evaluation protocol=Linear classification2024.03 | 87.8 | |
| EGMAPortion of training data=1%, Evaluation protocol=Linear classification2024.03 | 87.71 | |
| MGCAPortion of training data=10%, Evaluation protocol=Linear classification2024.03 | 87.66 | |
| MedCLIPPortion of training data=10%, Evaluation protocol=Linear classification2024.03 | 87.49 | |
| ConVIRTPortion of training data=100%, Evaluation protocol=Linear classification2024.03 | 87.3 | |
| PRIORPortion of training data=10%, Evaluation protocol=Linear classification2024.03 | 87.08 | |
| ConVIRTPortion of training data=10%, Evaluation protocol=Linear classification2024.03 | 86.8 | |
| GLORIAPortion of training data=1%, Evaluation protocol=Linear classification2024.03 | 86.6 | |
| Vanilla FL (FedAvg)Privacy Guarantee=Low, Backbone=DenseNet1212025.12 | 86.5 | |
| zkFL-HealthPrivacy Guarantee=High, Backbone=DenseNet1212025.12 | 86.4 | |
| PRIORPortion of training data=1%, Evaluation protocol=Linear classification2024.03 | 86.16 | |
| ConVIRTPortion of training data=1%, Evaluation protocol=Linear classification2024.03 | 85.9 | |
| MGCAPortion of training data=1%, Evaluation protocol=Linear classification2024.03 | 85.8 | |
| MedCLIPPortion of training data=1%, Evaluation protocol=Linear classification2024.03 | 85.74 | |
| BioViLPortion of training data=10%, Evaluation protocol=Linear classification2024.03 | 85.37 | |
| BioViLPortion of training data=1%, Evaluation protocol=Linear classification2024.03 | 81.95 | |
| FL + Diff. PrivacyPrivacy Guarantee=High, Backbone=DenseNet121, epsilon=22025.12 | 76 |