Findings-Diagnosis on RadCoverage-VQA (internal held-out evaluation set)
70.1CXR AccuracyHR1.5
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| HR1.52026.07 | 70.1 | 64.8 | 70 | 67.7 | 65.3 | 66.6 | |
| HR12026.07 | 65.9 | 63.1 | 66.9 | 67.5 | 60 | 64.3 | |
| HR1.5+2026.07 | 50.2 | 47.6 | 36.5 | 43.5 | 47.9 | 46.6 | |
| MedGemma-4BParameters=4B2026.07 | 25.5 | 22.9 | 16 | 16.5 | 26.5 | 22.4 | |
| MedGemma-27BParameters=27B2026.07 | 17.4 | 15.7 | 10.7 | 13.1 | 14.1 | 15.3 |