Article Figures & Data
Figures
- FIGURE 1.
PRISMA flow diagram demonstrating selection strategy of research articles included in review.
- FIGURE 2.
Categorywise split of selected articles reviewed here.
- FIGURE 3.
Typical supervised framework for PET image denoising using deep learning with training phase that minimizes loss function and validation phase that evaluates deep-learning model’s performance. conv = convolution; ReLU = rectified linear unit; maxpool = maximum pooling; up-conv = upsampling convolution.
Tables
Paper Data details Architecture Loss function Xiang et al. (21) 16 brain 18F-FDG PET + T1 MRI Autocontext CNN MSE Xu et al. (22) 9 brain 18F-FDG PET + T1 MRI U-Net MAE Wang et al. (23) 16 brain 18F-FDG PET Concatenated 3D cGAN with a 3D U-Netlike generator MAE; cGAN Ouyang et al. (24) 39 brain 18F-florbetaben PET cGAN with U-Net-like generator MAE; cGAN; perceptual; content Wang et al. (25) 20 simulated + 16 clinical brain 18F-FDG PET, T1 MRI, diffusion tensor imaging Locality adaptive multimodality GAN MAE; adversarial Schaefferkoetter et al. (26) 31 lung 18F-FDG PET U-Net MSE Spuhler et al. (27) 35 brain 18F-FDG PET Dilated U-Net MAE Xue et al. (28) 10 whole-body 18F-FDG PET 3D attention residual least-squares GAN MSE; least-squares adversarial Zhao et al. (29) 109 brain 18F-FDG PET/CT Supervised cycleGAN Adversarial; cycle consistency; identity Gong et al. (30) 9 cardiac torso 18F-FDG PET Parameter-transferred Wasserstein GAN MSE; Adversarial Gong et al. (31) 120 brain 18F-FDG PET; 140 brain 18F-MK-6240 PET + T1 MRI Denoising diffusion probabilistic model MSE Zhang et al. (32) 70 brain 18F-FDG PET + T1 MRI Spatially adaptive and transformer fusion network MAE Cui et al. (34) 10 lung 68Ga-PRGD2 PET/CT; 30 lung 18F-FDG PET/T1 MRI Modified 3D U-Net with DIP MSE Cui et al. (35) 10 lung 68Ga-PRGD2 PET/CT; 30 lung 18F-FDG PET/T1 MRI Modified 3D U-Net with DIP MSE Song et al. (36) 20 simulated brain 18F-FDG PET; 17 brain 18F-FDG PET + T1 MRI Noise2Void (U-Net with 3 resolution levels) MSE Liu et al. (37) 195 cardiac torso 18F-FDG PET 3D U-Net MSE Zhou et al. (38) Heterogeneous multiinstitutional PET Deep attention residual U-Net MSE Xue et al. (39) 310 brain across 18F-FDG, 18F-FET, 18F-florbetapir PET cGAN Conventional content; voxelwise Jang et al. (33) 44 whole-body 18F-FDG, 40 whole-body 18F-ACBC, 10 whole-body 18F-DCFPyL, 18 whole-body 68Ga-DOTATATE Spach transformer Charbonnier Song et al. (40) 20 simulated brain 18F-FDG PET; 30 clinical brain 18F-FDG PET + T1 MRI Very deep superresolution CNN MAE Song et al. (41) 20 simulated brain 18F-FDG PET; 30 clinical brain 18F-FDG PET Self-supervised superresolution (CycleGAN) Adversarial; cycle consistency; total variation Sanaat et al. (42) 50 brain 18F-FDG, 50 brain 18F-flortaucipir, 36 brain 18F-flutemetamol, 76 brain 18F-fluoro-DOPA + T1 MRI CycleGAN Adversarial Sanaat et al. (43) 100 brain 18F-FDG, 100 18F-flortaucipir, 100 brain 18F-flutemetamol CycleGAN Adversarial Azimi et al. (44) 160 brain 18F-FDG PET/CT Attention-based network (U-Net) MSE Mehranian et al. (45) 273 whole-body 18F-FDG PET 3D residual U-Net MSE MSE = mean-squared error; MAE = mean absolute error; 18F-ACBC = 1-amino-3-18F-fluorocyclobutane-1-carboxylic acid.
Paper Data details Architecture Loss function Ramon et al. (46) 930 cardiac torso 99mTc-sestamibi 3D convolutional autoencoder MSE Ramon et al. (47) 1,052 cardiac torso 99mTc-sestamibi Convolutional autoencoder, CNN MSE Sun et al. (48) 100 simulated; 20 cardiac torso clinical 99mTc-sestamibi Pix2Pix GAN MAE; adversarial Sohlberg et al. (49) 93 cardiac torso 99mTc-tetrososmin CNN, residual network, U-Net, cGAN MSE Yu et al. (50) 4,800 simulated CNN MSE Shiri et al. (51) 363 cardiac torso 99mTc-sestamibi Deep residual neural network MSE Lin et al. (52) 112 cardiac torso 99mTc-DMSA 3D residual U-Net MSE Pan et al. (53) 20 cardiac torso 99mTc-MDP SPECT/CT Lesion-attention weighted U2Net MAE; structural similarity index Liu et al. (54) 895 cardiac torso 99mTc-sestamibi Noise2Noise (U-Net) MSE Liu et al. (55) 1,050 cardiac torso 99mTc-sestamibi 3D-coupled UNet MSE Xie et al. (56) 28 cardiac 99mTc-RBC Densely connected multidimensional dynamic U-Net MAE; structural similarity index; Sobel operator; intramyocardial blood volume MSE = mean-squared error; MAE = mean absolute error; DMSA = pentavalent dimercaptosuccinic acid; MDP = methyl diphosphonate; RBC = red blood cell.
