# Generative Adversarial Networks

4 related articles

Subscribe to updates

arXiv Oct 25, 2026
CT-to-MRI synthesis without 3D diffusion: StaX-Net shares interleaved 2D blocks and reaches an abdominal perceptual distance of 642.1, below Pix2Pix
Cross-modality medical image translation has been troubled by anatomically inconsistent synthesis, and standard generative adversarial networks update generator and discriminator together, letting opposing gradients pull on shared representations. StaX-Net interleaves multi-scale computational blocks between the generator and discriminator, lets the discriminator update first while shared blocks are frozen during the generator step, and uses target MRI only during training to guide a feature transform. On the Abdomen/Pelvis, Head and Neck, and Thorax subsets of SynthRAD 2025 it is compared with five baselines, reaching an Abdomen/Pelvis VGG perceptual distance of 642.1, below Pix2Pix at 991.2 and UNet-AE at 1687.9.
Cross-modality medical image translation has been troubled by anatomically inconsistent synthesis, and standard generative adversarial networks update generator and discriminator together, letting opposing gradients pull on shared representations. StaX-Net interleaves multi-scale computational blocks between the generator and discriminator, lets the discriminator update first while shared blocks are frozen during the generator step, and uses target MRI only during training to guide a feature transform. On the Abdomen/Pelvis, Head and Neck, and Thorax subsets of SynthRAD 2025 it is compared with five baselines, reaching an Abdomen/Pelvis VGG perceptual distance of 642.1, below Pix2Pix at 991.2 and UNet-AE at 1687.9.
VinUniversity
Amin Golnari（First author）
StaX-Net Cross-Modality Medical Image Translation Generative Adversarial Networks Stackelberg-Inspired Asymmetric Optimization
medRxiv Oct 24, 2026
A standard ECG can now yield synthetic heart-muscle motion: synthetic e′ predicts post-discharge death on its own and lifts discrimination from 0.67 to 0.70
A multi-cohort external validation shows that a previously trained generative adversarial network, applied without retraining, can synthesize mitral annular tissue-Doppler velocities and displacement from a standard 12-lead ECG. Among 1,065 hospitalized acute myocardial infarction patients, synthetic e′ independently predicted all-cause mortality beyond the GRACE score; in the post-discharge subgroup it raised discrimination from a C-index of 0.67 to 0.70 and stayed independent of AI-ECG-predicted and measured ejection fraction. In 219,564 outpatients, all three parameters were associated with MI-specific mortality, with MAPSE showing the largest effect.
A multi-cohort external validation shows that a previously trained generative adversarial network, applied without retraining, can synthesize mitral annular tissue-Doppler velocities and displacement from a standard 12-lead ECG. Among 1,065 hospitalized acute myocardial infarction patients, synthetic e′ independently predicted all-cause mortality beyond the GRACE score; in the post-discharge subgroup it raised discrimination from a C-index of 0.67 to 0.70 and stayed independent of AI-ECG-predicted and measured ejection fraction. In 219,564 outpatients, all three parameters were associated with MI-specific mortality, with MAPSE showing the largest effect.
Synthetic Tissue Doppler Imaging Generative Adversarial Networks Mortality Prediction cardiovascular risk stratification
发表出处待核验 Sep 30, 2026
Estimating SV2A PET-Derived Synaptic Density from Quantitative MRI: 3D U-Net Reaches Pearson Correlation of 0.8838 in Gray Matter
Combining two multimodal qMRI-PET datasets (n = 74, spanning Alzheimer's disease, subjective cognitive decline, and healthy controls), this study used [18F]UCB-H PET distribution volume VT from Logan graphical analysis as the synaptic-density reference, applied ComBat harmonization, and compared classical machine learning (SVR, PLS, Elastic Net, Random Forests) with deep learning (U-Net, ResUNet++, Pix2Pix-like conditional GANs) for predicting PET-like synaptic density images from qMRI maps such as R1, R2*, MTsat, and PD; Elastic Net was best among classical models (R² = 0.50, RMSE = 0.448, MAE = 0.331), deep learning improved accuracy with 3D U-Net most consistent, and gray-matter z-scored evaluation showed strong agreement with reference PET (MSE 0.1294 ± 0.0778, SSIM 0.9832 ± 0.
Combining two multimodal qMRI-PET datasets (n = 74, spanning Alzheimer's disease, subjective cognitive decline, and healthy controls), this study used [18F]UCB-H PET distribution volume VT from Logan graphical analysis as the synaptic-density reference, applied ComBat harmonization, and compared classical machine learning (SVR, PLS, Elastic Net, Random Forests) with deep learning (U-Net, ResUNet++, Pix2Pix-like conditional GANs) for predicting PET-like synaptic density images from qMRI maps such as R1, R2*, MTsat, and PD; Elastic Net was best among classical models (R² = 0.50, RMSE = 0.448, MAE = 0.331), deep learning improved accuracy with 3D U-Net most consistent, and gray-matter z-scored evaluation showed strong agreement with reference PET (MSE 0.1294 ± 0.0778, SSIM 0.9832 ± 0.
University of Liège Haute École de la Province de Liège
Axel Ruymaekers（First author）
Synaptic Density Estimation from Quantitative MRI U-Net Generative Adversarial Networks Quantitative MRI
Acta Scientiae Sep 30, 2026
Survey reports that AI combining behavioral, genetic, and imaging data with GAN-based augmentation may improve autism spectrum disorder screening, but limited data, class imbalance, and scarce external clinical validation remain
This survey reviews AI, machine learning, deep learning, and generative adversarial network (GAN) approaches to autism spectrum disorder (ASD) screening, focusing on multimodal learning across behavioral, genetic, environmental, neuroimaging, physiological, and clinical data and on the role of GANs in synthetic-data generation and augmentation, concluding that multimodal AI may represent ASD-related characteristics more comprehensively than single-modality approaches while facing challenges of limited and heterogeneous datasets, class imbalance, multimodal integration, GAN training instability, synthetic-data quality, privacy and security, interpretability, generalizability, and limited external clinical validation.
This survey reviews AI, machine learning, deep learning, and generative adversarial network (GAN) approaches to autism spectrum disorder (ASD) screening, focusing on multimodal learning across behavioral, genetic, environmental, neuroimaging, physiological, and clinical data and on the role of GANs in synthetic-data generation and augmentation, concluding that multimodal AI may represent ASD-related characteristics more comprehensively than single-modality approaches while facing challenges of limited and heterogeneous datasets, class imbalance, multimodal integration, GAN training instability, synthetic-data quality, privacy and security, interpretability, generalizability, and limited external clinical validation.
Dr. Manjunath B N（First author）
Generative Adversarial Networks Autism Spectrum Disorder Screening Multimodal Fusion Data Augmentation

You’re all caught up

CT-to-MRI synthesis without 3D diffusion: StaX-Net shares interleaved 2D blocks and reaches an abdominal perceptual distance of 642.1, below Pix2Pix

A standard ECG can now yield synthetic heart-muscle motion: synthetic e′ predicts post-discharge death on its own and lifts discrimination from 0.67 to 0.70

Estimating SV2A PET-Derived Synaptic Density from Quantitative MRI: 3D U-Net Reaches Pearson Correlation of 0.8838 in Gray Matter

Survey reports that AI combining behavioral, genetic, and imaging data with GAN-based augmentation may improve autism spectrum disorder screening, but limited data, class imbalance, and scarce external clinical validation remain