Dual-Domain Equivariant Generative Adversarial Network for Multimodal CT-PET Synthesis

The Dual-Domain Equivariant Generative Adversarial Network (DDE-GAN) has been introduced for multimodal CT-PET image synthesis, addressing limitations of traditional GANs by leveraging both spatial and frequency domains to enhance structural fidelity. The model incorporates rotational equivariance into its loss functions to improve anatomical accuracy and employs a hierarchical dual-domain training strategy for consistency across domains. Evaluated on the HECKTOR 2022 dataset, DDE-GAN outperforms baseline models, presenting significant implications for practitioners in medical imaging and data augmentation applications.