Learning Disentangled Representations in the Imaging Domain