I have from multiple people annotation as masks of the same images.

How can I combine these masks to get a more “true” annotation? Is there some kind of standard?

I also thought of using things like Cohen’s kappa, Krippendorf’s alpha, F1 etc. but I just realized I can’t use those for quality assurance, as they are more for labelling than annotation in segmentation.
Do you know something I could also use for quality assurance?