Abnormality labeling in CT reports on CT-RATE 100 randomly sampled reports
96.4SensitivityMajority vote (3 LLMs)
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Majority vote (3 LLMs)Aggregation=Multi-LLM majority-vote2026.06 | 96.4 | 98.8 | 95.7 | 99.3 | 98.6 | 96 | 0.951 | |
| CT-RATE (original)Label Source=Original labels2026.06 | 96.3 | 97.7 | 92.9 | 99.2 | 97.6 | 93.8 | 0.923 | |
| DeepSeek-V3.2Model=DeepSeek-V3.2, Configuration=Individual LLM-derived labels2026.06 | 95.5 | 98.6 | 94 | 98.7 | 97.9 | 94.5 | 0.931 | |
| GPT-5.4 miniModel=GPT-5.4 mini, Configuration=Individual LLM-derived labels2026.06 | 94.8 | 98.5 | 94.9 | 99 | 98.2 | 94.6 | 0.934 | |
| GPT-5.4Model=GPT-5.4, Configuration=Individual LLM-derived labels2026.06 | 94.4 | 98.8 | 96.8 | 98.9 | 98.4 | 95.1 | 0.94 |