Text this: Comparison of synthetic LGE with optimal inversion time vs. conventional LGE via representation learning