Medical Report Generation on CT-RATE-Report full (test)
41.24PrecisionOurs w/ TIF-GRPO
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Ours w/ TIF-GRPO#Params=8B2026.05 | 41.24 | 38.21 | 39.67 | 87.66 | 71.06 | 64.4 | 20.67 | 40.51 | 13.57 | 41.04 | 70.26 | 83.67 | |
| Ours w/ TIF-GRPO#Params=4B2026.05 | 40.93 | 37.8 | 39.11 | 87.28 | 70.23 | 65.79 | 20.37 | 40.84 | 12.69 | 42.18 | 70.86 | 83.46 | |
| GPT-5#Params=*2026.05 | 39.71 | 19.94 | 26.26 | 73.79 | 69.87 | 47.28 | 8.33 | 18.48 | 5.09 | 13.08 | 50.73 | 74.15 | |
| Ours w/ SFT#Params=4B2026.05 | 34.59 | 12.85 | 18.73 | 61.45 | 67.99 | 52.22 | 5.68 | 12.06 | 2.71 | 37.09 | 66.88 | 82.26 | |
| Gemini-3-Pro#Params=*2026.05 | 24.58 | 18.96 | 21.37 | 81.46 | 63.71 | 41.06 | 6.26 | 17.85 | 3.93 | 17.1 | 54.91 | 75.02 | |
| CT-CHAT†#Params=8B2026.05 | 23.91 | 29.24 | 26.03 | 87.53 | 78.04 | 51.67 | 14.98 | 32.43 | 10.34 | 31.23 | 64.53 | 80.52 | |
| InternVL3.5-8B#Params=8B2026.05 | 23.62 | 4.61 | 7.7 | 37.92 | 55.63 | 30.41 | 0.74 | 3.33 | 0.18 | 16.95 | 54.6 | 74.83 | |
| Fleming-VL-8B#Params=8B2026.05 | 20.52 | 2.34 | 4.18 | 29.22 | 49.29 | 24.63 | 0.47 | 1.56 | 0.27 | 16.26 | 55.33 | 73.91 | |
| Hulu-Med#Params=7B2026.05 | 20.45 | 10.16 | 12.53 | 47.31 | 61.45 | 33.58 | 3.19 | 9.81 | 1.98 | 13.96 | 53.44 | 73.37 | |
| Qwen3-VL-8B#Params=8B2026.05 | 13.73 | 9.8 | 11.42 | 84.14 | 56.18 | 29.75 | 3.08 | 11.66 | 1.75 | 16.06 | 52.74 | 73.9 | |
| RadFM#Params=14B2026.05 | 13.29 | 4.15 | 6.3 | 36 | 41.18 | 30.36 | 0.52 | 3.77 | 0.3 | 7.1 | 35.5 | 68.74 | |
| M3D-LaMed#Params=7B2026.05 | 8.77 | 5.98 | 7.02 | 85.17 | 35.56 | 18.76 | 0.56 | 6.47 | — | 8.39 | 38.74 | 69.58 |