Projection Image Synthesis Using Adversarial Learning Based Spatial Transformer Network For Sparse Angle Sampling CT

In sparse-angle X-ray tomography, CT images possess significant noise and artifacts in reconstruction. An important image reconstruction research effort in this area aims to remove these artifacts while preserving image features. Approaches to the problem can be divided into two main categories, post-processing on reconstructed images and pre-processing on the sinogram. Following the rise of deep learning methods, many data-driven models have been proposed in both categories. However, those methods require a large amount of fully sampled data for training. In this paper, inspired by the development of the video frame synthesis technique, we propose an adversarial learning-based spatial transformer network for projection image synthesis, which aims to improve the quality of reconstructed images by reasonably increasing the number of projection images based on the existing projection data. Simulation and experiments have shown that this data-driven model can give rise to competitive results compared with conventional algorithms.