Synthetic PET

Synthetic PET image generation leverages deep learning to create realistic positron emission tomography (PET) scans from readily available modalities like MRI, addressing the limitations of PET's high cost and radiation exposure. Current research focuses on developing advanced generative models, including diffusion models and U-Nets, to accurately translate structural MRI information into functional PET data, often incorporating strategies to improve image fidelity and preserve pathological details. This capability is significantly impacting medical image analysis by enabling data augmentation for improved diagnostic model training, facilitating research on PET reconstruction algorithms without the need for extensive patient datasets, and potentially expanding access to PET-based diagnostics.