A method, computer system, and a computer program product for generating one or more realistic x-ray angiography (XA) images for deep learning is provided. The present invention may include generating at least one three-dimensional (3D) model associated with a subject. The present invention may then include projecting the generated 3D model associated with the subject into one or more two-dimensional (2D) images associated with the subject. The present invention may also include creating the one or more realistic XA images of the subject from the one or more 2D images associated with the subject by utilizing a trained deep convolutional neural network.