I have a question — this paper which showed that MC dropout doesn’t actually estimates the true posterior distribution.

https://www.google.com/url?sa=t&source=web&rct=j&url=https://arxiv.org/abs/1711.02989&ved=2ahUKEwjjudemsMDpAhW9wzgGHU0sCjIQFjAAegQIAxAC&usg=AOvVaw1RHXgYCtA7XLo3T7lpBfUv

Is mc dropout reliable in extremely sensitive situation like healthcare?