Visual Reasoning on DeepTumorVQA refined 2025b
60Adjacent Organ AccuracyMedGemma
Evaluation Results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MedGemmaParameters=27B, Evaluation Protocol=zero-shot2026.02 | 60 | 40 | 37 | 37 | 28 | 37 | 23 | 32 | 62 | 47 | 37 | 22 | 45 | 30 | 38 | |
| HuatuoGPTParameters=34B, Evaluation Protocol=zero-shot2026.02 | 60 | 50 | 35 | 33 | 22 | 53 | 27 | 22 | 48 | 43 | 35 | 17 | 42 | 20 | 36 | |
| RadFMEvaluation Protocol=zero-shot2026.02 | 55 | 23 | 38 | 32 | 10 | 43 | 15 | 18 | 45 | 55 | 33 | 14 | 38 | 32 | 34 | |
| M3DEvaluation Protocol=zero-shot2026.02 | 53 | 23 | 32 | 30 | 12 | 53 | 15 | 20 | 44 | 47 | 40 | 24 | 53 | 23 | 34 | |
| Qwen3-VLParameters=30B, Evaluation Protocol=zero-shot2026.02 | 45 | 30 | 27 | 38 | 17 | 52 | 20 | 43 | 56 | 50 | 42 | 27 | 37 | 23 | 35 | |
| 3DMedAgent (GPT-5)Base Model=GPT-5, Evaluation Protocol=zero-shot2026.02 | 45 | 53 | 70 | 45 | 43 | 60 | 52 | 45 | 66 | 65 | 43 | 55 | 88 | 52 | 56 | |
| GPT-5Evaluation Protocol=zero-shot2026.02 | 43 | 38 | 30 | 30 | 32 | 40 | 40 | 42 | 39 | 50 | 42 | 36 | 50 | 40 | 39 | |
| 3DMedAgent (Qwen3-VL)Base Model=Qwen3-VL (30B), Evaluation Protocol=zero-shot2026.02 | 42 | 45 | 62 | 47 | 40 | 55 | 43 | 48 | 65 | 55 | 38 | 58 | 82 | 47 | 52 | |
| RandType=Random baseline2026.02 | 33 | 33 | 33 | 33 | 25 | 42 | 25 | 25 | 33 | 50 | 33 | 25 | 50 | 28 | 34 |