Medical Report Generation on MIMIC-CXR (NLG, Clinical, and Custom Metrics)
0.206BLEU-2RadHiera
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| RadHieraModel=RadHiera2025.11 | 0.206 | 0.421 | 0.442 | 0.282 | 0.939 | 39.4 | 0.851 | — | |
| R2GenGPTModel=R2GenGPT2025.11 | 0.203 | 0.407 | 0.305 | 0.243 | 0.909 | 34.8 | 0.753 | — | |
| CheXpertModel=CheXpert2025.11 | 0.196 | 0.389 | 0.429 | 0.166 | 0.838 | 32.2 | 0.729 | — | |
| MedVersaModel=MedVersa2025.11 | 0.193 | 0.43 | 0.315 | 0.273 | 0.919 | 37.6 | 0.776 | — | |
| MedGemma-4BModel=MedGemma-4B2025.11 | 0.185 | 0.417 | 0.301 | 0.258 | 0.905 | 35.9 | 0.771 | — | |
| RadFMModel=RadFM2025.11 | 0.081 | 0.281 | 0.245 | 0.111 | 0.625 | 24.5 | 0.741 | — | |
| MedGemma 4BEvaluation Mode=Out-of-box2025.07 | — | — | — | — | — | — | — | 29.5 | |
| MedGemma 4BEvaluation Mode=Fine-tuned2025.07 | — | — | — | — | — | — | — | 30.3 | |
| MedVersa2025.07 | — | — | — | — | — | — | — | 30 |