Positron emission tomography image enhancement using magnetic resonance images and U-net structure

•A Single image super resolution based on deep learning produces high quality PET images•Exploiting MRI images has an important role in intensifying the quality of PET images•Combination of U-Net structure with residual blocks improves the system performance Positron Emission Tomography (PET) has become an important tool for diagnosing abnormalities, but it suffers from low spatial resolution and a high-level noise. In this article, a Convolutional Neural Network (CNN)-based Single Image Super-resolution (SISR) method is used to produce a PET image with a desired quality. The T1-Weighted Magnetic Resonance (MR) images are used to enrich the information applied to the network. A network based on U-Net structure is used and residual blocks are inserted into the network to improve system performance. This article also evaluates the impact of various loss functions, such as Mean Squared Error (MSE) and its combination with a perceptual loss on the efficiency of the proposed method. Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index (SSIM) on two various databases (simulated and clinical data) are 36.78, 0.9927, and 37.36, 0.9714, respectively, indicating good performance of the proposed method compared to previous works. [Display omitted]