Performance of a machine-learning algorithm for fully automatic LGE scar quantification in the large multi-national derivate registry