Dual-Domain Equivariant Generative Adversarial Network for Multimodal CT-PET Synthesis
Researchers have developed a Dual-Domain Equivariant Generative Adversarial Network (DDE-GAN) to improve the synthesis of multimodal CT-PET images. Unlike traditional methods that focus only on spatial domains, DDE-GAN incorporates both spatial and frequency (Fourier) domains to capture richer anatomical and spectral information. The network also integrates rotational equivariance, ensuring consistent responses under rotation and enhancing anatomical accuracy. Tested on the HECKTOR 2022 dataset, DDE-GAN demonstrated superior synthesis quality, offering potential for PET completion and data augmentation. AI
IMPACT This research could lead to more accurate and robust multimodal image synthesis, improving applications in medical imaging and data augmentation.