SPC-Gen

SPC-Gen is a large-scale, physically consistent synthetic PET/CT dataset designed to address the scarcity of medical imaging data, particularly in nuclear medicine. The dataset is generated using a latent diffusion model trained directly in the projection domain, thereby preserving essential physical information of the PET imaging process, including photon counting statistics and system response characteristics.SPC-Gen comprises over 50,000 paired slices in both the projection and image domains, covering a wide range of tracers (e.g., ¹⁸F-FDG, ¹⁸F-DOPA, ⁶⁸Ga-PSMA), anatomical regions (brain, trunk, abdomen, chest), and imaging modalities (PET and CT). The dataset is organized to support both single-channel and multi-channel configurations (e.g., dual-tracer fusion), making it suitable for various downstream tasks such as tracer separation, image fusion, lesion segmentation, and low-dose CT reconstruction.All generated data have undergone rigorous validation to ensure distributional consistency, physical authenticity, and task-level usability. Data are provided in multiple formats, including DICOM, NumPy (.npy), and MATLAB (.mat), accompanied by comprehensive metadata to facilitate flexible data filtering and experimental design.By releasing SPC-Gen, we aim to provide a standardized benchmark resource for PET/CT imaging research, generative model evaluation, and data augmentation, ultimately supporting the development of physically grounded and clinically meaningful AI models in medical imaging.