Presence detection on CheXpert direct presence (n=235)
74.04Base AccuracyLLaVA-Med
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LLaVA-MedChain-of-Thought (CoT)=false2026.05 | 74.04 | 74.04 | 0 | 0 | |
| LLaVA-MedChain-of-Thought (CoT)=true2026.05 | 68.09 | 68.09 | 0 | 0 | |
| GPT-4oChain-of-Thought (CoT)=false2026.05 | 62.55 | 73.19 | 10.64 | 25 | |
| LLaVA-1.5Chain-of-Thought (CoT)=false2026.05 | 56.6 | 68.09 | 11.49 | 27 | |
| Qwen2.5-VLChain-of-Thought (CoT)=true2026.05 | 54.47 | 78.72 | 24.26 | 57 | |
| MedGemmaModel size=27B, Chain-of-Thought (CoT)=false2026.05 | 51.91 | 96.6 | 44.68 | 105 | |
| MedGemmaChain-of-Thought (CoT)=true2026.05 | 42.98 | 94.89 | 51.91 | 122 | |
| CheXagentChain-of-Thought (CoT)=false2026.05 | 37.87 | 40.85 | 2.98 | 7 | |
| MedGemmaChain-of-Thought (CoT)=false2026.05 | 31.49 | 96.6 | 65.11 | 153 | |
| Qwen2.5-VLChain-of-Thought (CoT)=false2026.05 | 30.21 | 95.32 | 65.11 | 153 |