Learning a Deep Multi-Scale Feature Ensemble and an Edge-Attention Guidance for Image Fusion | Synapse