Fig. 8: Correlation between automatic NLP metrics and human reader preference scores. | npj Digital Medicine