Department of Nuclear Medicine, Ulsan University Hospital, Ulsan 44033, Republic of Korea.
Department of Nuclear Medicine, Ulsan University Hospital, University of Ulsan College of Medicine, Ulsan 44033, Republic of Korea.
Medicina (Kaunas). 2023 Jul 10;59(7):1281. doi: 10.3390/medicina59071281.
: F-fluorodeoxyglucose (FDG) positron emission tomography (PET) (PET) image can visualize neuronal injury of the brain in Alzheimer's disease. Early-phase amyloid PET image is reported to be similar to PET image. This study aimed to generate PET images from F-florbetaben PET (PET) images using a generative adversarial network (GAN) and compare the generated PET (PET) with real PET (PET) images using the structural similarity index measure (SSIM) and the peak signal-to-noise ratio (PSNR). : Using the Alzheimer's Disease Neuroimaging Initiative (ADNI) database, 110 participants with both PET and PET images at baseline were included. The paired PET and PET 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 and PET 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 and PET images in the cycleGAN and pix2pix models were evaluated using the independent Student's -test. Statistical significance was set at ≤ 0.05. : 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 and PET 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 images of the cycleGAN model showed statistically higher mean SSIM than those of the pix2pix model ( < 0.001). The mean PSNR was also higher in the PET images of the cycleGAN model than those of pix2pix model ( < 0.001). : We generated PET images from PET images using deep learning. The cycleGAN model generated PET 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 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.
正电子发射断层扫描（PET）图像可以可视化阿尔茨海默病患者大脑的神经元损伤。早期淀粉样蛋白 PET 图像报告与 PET 图像相似。本研究旨在使用生成对抗网络（GAN）从 F-氟脱氧葡萄糖（FDG）PET（PET）图像生成 PET（PET）图像，并使用结构相似性指数测量（SSIM）和峰值信噪比（PSNR）比较生成的 PET（PET）与真实 PET（PET）图像。
使用阿尔茨海默病神经影像学倡议（ADNI）数据库，纳入了 110 名基线时既有 PET 又有 PET 图像的参与者。配对的 PET 和 PET 图像分别包含 6 个和 4 个子集图像。每个子集图像的采集时间为 5 分钟。这些子集被随机抽样并分为 249 对 PET 和 PET 子集图像，用于深度学习过程中的训练数据集和 95 对子集图像，用于验证数据集。本研究中使用的深度学习模型由具有 U-Net 的 GAN 组成。使用独立学生 t 检验评估 cycleGAN 和 pix2pix 模型中 PET 和 PET 图像的 SSIM 和 PSNR 值之间的差异。统计显著性设置为 ≤ 0.05。
参与者的人口统计学数据（年龄、性别或诊断）在训练组（82 名参与者）和验证组（28 名参与者）之间没有统计学上的显著差异。cycleGAN 模型的 PET 和 PET 图像之间的平均 SSIM 为 0.768 ± 0.135，pix2pix 模型的平均 SSIM 为 0.745 ± 0.143。平均 PSNR 分别为 32.4 ± 9.5 和 30.7 ± 8.0。cycleGAN 模型的 PET 图像的平均 SSIM 显著高于 pix2pix 模型（ < 0.001）。cycleGAN 模型的 PET 图像的平均 PSNR也高于 pix2pix 模型（ < 0.001）。
我们使用深度学习从 PET 图像生成 PET 图像。cycleGAN 模型生成的 PET 图像具有更高的 SSIM 和 PSNR 值。使用深度学习进行图像到图像的转换可能有助于生成 PET 图像。这些图像可能无需额外的图像采集以及随之而来的辐射暴露、不便或费用增加，为阿尔茨海默病的管理提供额外信息。
