List your company

Radiology report discrepancy evaluation on RadEvalX 100-pair

0.715Spearman Correlation (Total)RadOT-Eval
0.083720.247610.41150.57539Jun 7, 2026

Evaluation results

MethodLinks
RadOT-Evalvariant=full2026.060.7150.5480.399
RadOT-Evalvariant=w/o ref count2026.060.7120.5480.308
Structured monotone2026.060.6120.3980.256
GNN discrepancy model2026.060.6070.3980.119
Text-only ridge2026.060.5920.3370.324
GREEN risk2026.060.570.4330.141
Graph-derived monotone2026.060.5690.4690.206
CheXbert2026.060.4920.4130.103
Reference count only2026.060.4740.2010.41
BERTScore2026.060.3490.1950.151
RadCliQ2026.060.3350.1880.177
GREEN native count2026.060.2960.382—
BLEU-22026.060.2920.1620.157
RadGraph-F12026.060.2850.1590.147
BLEU-42026.060.1080.0960.046