A long-standing challenge in developing deep learning approaches has been the lack of high-quality datasets, especially in material and medical. Different from the lack of the scanning transmission electron microscope (STEM) images dataset in material, mammogram images can not to be simulated completely using mathematic or physics theory. Doctors can only diagnose the breast cancer in early detection with limited data. The scarcity and imbalance of mammogram dataset is the main reason led to various performance problems in computer-aided diagnosis (CAD) methods. Generate more high-quality mammograms is an immediate need of time. However, the wider augmentation of mammogram dataset has been hindered by the lack of method that synthesis high-resolution mammograms efficiently. To overcome this problem, we proposed an edge-guided progressive growing generative adversarial network (EPG-GAN) to efficiently synthesize high-resolution mammograms. By introducing the design of edge-guided into the EPG-GAN, the model is empowered to be able to avoid synthesizing unnatural images and thus generate mammograms effectively and more realistic. Turning progressive growing strategy as global training strategy, this design leverages the fusion block to fade the image trained by previous step in the next step smoothly and make sure that EPG-GAN can gradually generate mammograms from low to high resolution. To demonstrate the effectiveness of our proposed model, we conduct experiments on publicly available INbreast dataset. Qualitative and quantitative evaluations validate the performance of our model in synthesis high-resolution mammogram. The proposed EPG-GAN achieves 0.983 in structure similarity index measure (SSIM) score, which is significantly higher than other models. The results show that the EPG-GAN can effectively synthesis realistic and high-resolution mammograms, enabling the augmentation of the dataset and laying the foundation for enhancing CAD for breast cancer.