Weight-Adaptive Network With CT Enhancement for Short-Duration PET Imaging Utilizing the uEXPLORER Total-Body System

Abstract

The total-body positron emission tomography (PET) scanning time is typically reduced to mitigate motion artifacts, yet this can compromise image quality. Current approaches often enhance PET resolution via CT guidance but overlook structural disparities across anatomical sites. Therefore, this article introduces an enhanced Wasserstein generative adversarial network with gradient penalty (WGAN-GP), integrating anatomical information as attributes to enhance quality of multiple short-duration (2.5%, 5%, and 10%) total-body PET images simultaneously. The proposed method is a weight-adaptive three-channel network for different regions, integrating PET/CT features and attributes to optimize short-duration PET image generation. peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), root mean square error (RMSE), and standard uptake value (SUV) are analyzed within whole images and regions of interests (ROIs) to compare proposed method with other networks. The results on the 18F-FDG PET dataset show the method achieves better-visual effects and metrics (like SSIM: 0.94±0.04 for 2.5%; 0.95±0.04 for 5%; and 0.96±0.04 for 10%) across total-body than others. Furthermore, the SUV-maximum and activity distributions of ROIs are closest to standard-duration PET. Additionally, the method demonstrates robustness under varying degrees of 18F-FDG PET/CT misalignment and in the PSMA PET/CT dataset. The proposed method offers reliable technical support for clinical diagnosis via short-duration total-body PET.