Demographic conditioned DDPM: pseudo-normal PET generation for improved epileptic lesion detection

Motivation: High quality and demographically-matched control groups are critical for lesion detection in neurological diseases such as epilepsy. However, for FDG PET, good controls are not always available. Goal(s): To generate individualized control images for FDG PET from structural MRI images，thus enhancing lesion detection for patients with focal epilepsy. Approach: We propose to leverage data from both healthy control and patients using improved DDPM-based model that incorporates demographic and clinical information to achieve precise MRI to PET mapping for healthy controls. Then we combine personalized pseudo healthy reference and patients' FDG PET to obtain lesion probability maps. Impact: Paired MRI and FDG PET brain images of normal subjects are scarce. We integrate controls' and patients' demographic information into MRI2PET image translation using improved DDPM-based model, which will provide personalized pseudo-normal PET reference to aid lesion detection.