3D Convolutional Neural Network to Enhance Small-Animal Positron Emission Tomography Images in the Sinogram Domain

Abstract

In this work, we propose a three dimensional (3D) convolutional neural network (CNN) to enhance sinograms acquired from a small-animal positron emission tomography (PET) scanner. The network consists of three convolutional layers created with 3D filters of sizes 9, 3, and 5, respectively. We extracted 15250 3D patches from low- and highcount sinograms to build the low- and high-resolution pairs for training. After training and prediction, the enhanced sinogram is reconstructed using the ordered subset expectation maximization (OSEM) algorithm. The results revealed that the proposed network improved the spillover ratio and the uniformity of the standard NU4-2008 phantom up to 8% and 75%, respectively.