Deep Learning Based Framework for Direct Reconstruction of PET Images

Abstract

In Positron Emission Tomography (PET), high radioactivity maps are essential to better understand the physiological processes associated with the disease. In this paper, we propose a deep learning based framework for PET image reconstruction from sinogram domain directly. In the framework, conditional Generative Adversarial Networks (cGANs) is constructed to learn a mapping from sinogram data to reconstructed image and generate a well-trained model. To verify the accuracy and robustness of the model, both Monte Carlo simulation data and real data are adopted in the test. The experimental results show that the proposed framework is of great robustness and the reconstructed image is much more accurate in comparison with the traditional methods.