Question regarding regression model (RadCliQ)

[rphlstck]

Hey,
first of all thank you for the great work!
I have a question regarding the regression model aka RadCliQ. If I understood correctly, RadCliQ is a regression model, which predicts the average number of errors a radiologist would assign to a generated report. The RadCliQ input parameters are the BLEU score and the RadGraph score, which are calculated given a generated report and a ground truth report. (Hopefully, I understood this part correctly.)

Now coming to the question: Why did you only provide these two scores to the regression model and not additionally the BERTScore as it appears to have a good correlation to the radiologists error voting as seen in figure 3 of your paper? (https://www.medrxiv.org/content/10.1101/2022.08.30.22279318v1.full.pdf)