Learning Interpretable Anatomical Features Through Deep Generative Models: Application to Cardiac Remodeling