We propose a generative adversarial approach to address the medical image synthesis problem. Specifically, we train a fully convolutional network (FCN) to generate a target image given a source image. To produce more realistic image, we propose to use adversarial learning strategy to better model the FCN. Moreover, an image-gradient-difference based loss function is proposed to avoid generating blurry images. Also, a long-term residual unit is explored to help the training of the network. We further apply auto-context model to implement a context-aware framework. Experimental results show the robustness and accuracy of our method in synthesizing various medical images.