Using StyleGANs for Visual Interpretability of Deep Learning Models on Medical Images