Researchers have developed a new method called HDDPM (Heteroscedastic Denoising Diffusion Probabilistic Model) to improve the recovery of low-count Positron Emission Tomography (PET) images. Unlike standard diffusion models that assume uniform noise, HDDPM accounts for the non-Gaussian and spatially dependent noise inherent in PET scans, particularly at low radiation doses. The model incorporates a Poisson-based variance module to generate voxel-wise noise maps, reflecting the activity-dependent noise structure. Evaluations across different scanners and dose levels showed HDDPM achieved comparable overall image quality to standard models but significantly reduced measurement errors in low-dose scenarios, demonstrating its reliability and physical motivation for quantitative PET recovery. AI
IMPACT This research could lead to more accurate medical imaging with lower radiation doses, improving diagnostic capabilities in healthcare.
RANK_REASON Research paper detailing a new model for image recovery. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
