Finding Extraction on MIMIC-CXR-EN 300-report (dev)
98.7Atelectasis AccuracyQwen3-14B
Evaluation Results
| Method | Links | |||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen3-14BPrompt=Optimized for this model2026.03 | 98.7 | 98 | 97.3 | 98 | 95.7 | 99.3 | 96.3 | 93 | 99 | 96.7 | 94.3 | 96.7 | 98.3 | 92 | 99.3 | 99.7 | 98 | 95.3 | 99 | 94.7 | 96 | |
| Qwen-Plus2026.03 | 98 | 98.7 | 98.3 | 97.7 | 98 | 99 | 97 | 96.3 | 98.7 | 99 | 95 | 95.7 | 100 | 95.3 | 98.3 | 99 | 99 | 97.7 | 99.3 | 95 | 95.7 | |
| GPT-4o-mini2026.03 | 98 | 98.7 | 97.7 | 98 | 82.3 | 99 | 88 | 94 | 97 | 90 | 93.7 | 95.7 | 100 | 93.3 | 99 | 99.7 | 99 | 97.7 | 99.3 | 96 | 96.3 | |
| DeepSeek-R12026.03 | 97.3 | 96.7 | 99 | 98.3 | 86.7 | 99 | 90.7 | 97 | 96.7 | 98.7 | 94 | 94.7 | 100 | 95.7 | 99.3 | 97.7 | 99.3 | 97.7 | 99 | 94.7 | 91.7 | |
| GPT-3.5-turbo2026.03 | 78.7 | 47.7 | 78.7 | 62.3 | 78 | 94.7 | 66.3 | 78.3 | 84.7 | 56.7 | 72 | 77.3 | 94.3 | 79.7 | 96.7 | 91.3 | 84.3 | 69 | 80.3 | 87.3 | 86 |