Skip to main content
. 2023 Aug 3;4(9):100802. doi: 10.1016/j.patter.2023.100802

Table 5.

RadCliQ scores of impression-generating models, including metric-oracle models, CXR-RePaiR, and the random retrieval baseline model

RadCliQ
BLEU metric-oracle 0.081
BERTScore metric-oracle −0.095
CheXbert metric-oracle 0.052
RadGraph metric-oracle −0.020
CXR-RePaiR 1.642
Random retrieval of impression 1.755

Lower is better. The 95% confidence interval and range of metric scores are available in Table S5.

indicates the best-performing model.