VQPET: Leveraging Vector-Quantized Codebook Prior for PET Image Synthesis
Published in IEEE Transactions on Medical Imaging, 45(5), 2542–2560, 2026
VQPET addresses PET image synthesis with a two-stage framework that learns a vector-quantized codebook prior from high-quality PET data and then uses that prior to guide downstream synthesis. The paper introduces a pretrained VQGAN-based prior and a codebook-prior-guided network to improve realism and robustness. The final article appears in volume 45, issue 5 of IEEE Transactions on Medical Imaging, pages 2542–2560.
This work aligns with my interest in building stronger representation priors for challenging medical imaging problems.
Related links:
Recommended citation: H. Chen, Z. Yang, Y. Zhou, X. Zhang, H. Zhang, D. Zhao, B. Wei, G. Zhou, and Y. Xu. (2026). "VQPET: Leveraging Vector-Quantized Codebook Prior for PET Image Synthesis." IEEE Transactions on Medical Imaging, 45(5), 2542–2560.
Download Paper