Accelerated Resolution Recovery Image Reconstruction Using a Neural Network Leapfrogging

Abstract

Point Spread Function (PSF) is widely modeled in iterative image reconstruction algorithms. However, PSF introduces Gibbs artifact in early stage and reduces it very slowly with further iterations. We have investigated application of neural network to image reconstruction with PSF to leapfrog the iteration gap. One neural network and experiments on brain PET data are presented. Results are promising: artifacts are reduced while quantification is maintained. Use of neural network provides us an alternate, fast way to address this computationally challenging issue.