Generation of Conventional (18)F-FDG PET Images from (18)F-Florbetaben PET Images Using Generative Adversarial Network: A Preliminary Study Using ADNI Dataset

Background and Objectives: (18)F-fluorodeoxyglucose (FDG) positron emission tomography (PET) (PET(FDG)) image can visualize neuronal injury of the brain in Alzheimer’s disease. Early-phase amyloid PET image is reported to be similar to PET(FDG) image. This study aimed to generate PET(FDG) images from (18)F-florbetaben PET (PET(FBB)) images using a generative adversarial network (GAN) and compare the generated PET(FDG) (PET(GE-FDG)) with real PET(FDG) (PET(RE-FDG)) images using the structural similarity index measure (SSIM) and the peak signal-to-noise ratio (PSNR). Materials and Methods: Using the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database, 110 participants with both PET(FDG) and PET(FBB) images at baseline were included. The paired PET(FDG) and PET(FBB) images included six and four subset images, respectively. Each subset image had a 5 min acquisition time. These subsets were randomly sampled and divided into 249 paired PET(FDG) and PET(FBB) subset images for the training datasets and 95 paired subset images for the validation datasets during the deep-learning process. The deep learning model used in this study is composed of a GAN with a U-Net. The differences in the SSIM and PSNR values between the PET(GE-FDG) and PET(RE-FDG) images in the cycleGAN and pix2pix models were evaluated using the independent Student’s t-test. Statistical significance was set at p ≤ 0.05. Results: The participant demographics (age, sex, or diagnosis) showed no statistically significant differences between the training (82 participants) and validation (28 participants) groups. The mean SSIM between the PET(GE-FDG) and PET(RE-FDG) images was 0.768 ± 0.135 for the cycleGAN model and 0.745 ± 0.143 for the pix2pix model. The mean PSNR was 32.4 ± 9.5 and 30.7 ± 8.0. The PET(GE-FDG) images of the cycleGAN model showed statistically higher mean SSIM than those of the pix2pix model (p < 0.001). The mean PSNR was also higher in the PET(GE-FDG) images of the cycleGAN model than those of pix2pix model (p < 0.001). Conclusions: We generated PET(FDG) images from PET(FBB) images using deep learning. The cycleGAN model generated PET(GE-FDG) images with a higher SSIM and PSNR values than the pix2pix model. Image-to-image translation using deep learning may be useful for generating PET(FDG) images. These may provide additional information for the management of Alzheimer’s disease without extra image acquisition and the consequent increase in radiation exposure, inconvenience, or expenses.