Synthetic PET and CT Data using Generative Model in Projection Domain

Summary

The rapid advancement of artificial intelligence has substantially increased the demand for large-scale datasets; however, available datasets remain insufficient for data-intensive tasks, particularly in nuclear medicine. While large-scale datasets exist for computed tomography (CT), few provide paired projection-domain and image-domain data across diverse lesion types. This issue is more pronounced for positron emission tomography (PET), where datasets are scarce, particularly for multiple tracers such as 18F-DOPA, 18F-FDG, and 68Ga-PSMA, anatomical regions, and disease categories. This work presents SPC-Gen, a large-scale Synthetic dataset for PET and CT, Generated using a latent diffusion model. The dataset covers diverse disease types, lesion characteristics, tracers, anatomical regions, and modalities, while preserving the intrinsic correspondence between projection and image domains. SPC-Gen contains 50,104 released synthetic slice records, including PET projection-image records and CT image-domain records, forming a standardized large-scale dataset. Experimental results show that the generated samples closely align with real data distributions and can support improved performance on downstream tasks such as lesion segmentation. Compared with conventional VAE models and more recent GAN- and NCSN-based baselines, SPC-Gen achieved stable KID values around 0.021 across the single-modality evaluation and an FID of 40.960 for the PET Trunk image-domain setting, indicating improved global distributional alignment between synthetic and real data.