Hallucination Detection on HalluCXR finding level 1.0
0False Alarm RateSimple threshold voting
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Simple threshold votingThreshold (k)=≥52026.05 | 0 | 80.8 | 100 | 19.2 | 32.2 | |
| Simple threshold votingThreshold (k)=≥62026.05 | 0 | 85.1 | 100 | 14.9 | 25.9 | |
| Simple threshold votingThreshold (k)=≥42026.05 | 0.8 | 72.7 | 84.6 | 27.3 | 41.3 | |
| Label-aware weighted votingEvaluation=5-fold CV2026.05 | 5 | 51.9 | 59.9 | 48.1 | 53.4 | |
| Simple threshold votingThreshold (k)=≥32026.05 | 5.4 | 53 | 57.6 | 47 | 51.8 | |
| ECE-weighted votingThreshold=0.52026.05 | 9.9 | 35.8 | 50.1 | 64.2 | 56.3 | |
| Simple threshold votingThreshold (k)=≥22026.05 | 14.9 | 30 | 42.3 | 70 | 52.7 | |
| Simple threshold votingThreshold (k)=≥12026.05 | 32.9 | 12.4 | 29.3 | 87.6 | 43.9 |