• Lowering dose in SV2A PET impairs synaptic density measure in Parkinson’s disease (PD). • Two-step deep image prior (DIP) separates PD and healthy groups better in low-dose PET. • Two-step DIP strengthens correlation between synaptic density and motor severity. • Self-supervised deep learning enables dose reduction in dynamic PET imaging. Dynamic PET imaging with 11 C-UCB-J enables in vivo quantification of synaptic vesicle glycoprotein 2A (SV2A), with prior reports of lower synaptic density in areas such as the brainstem nuclei and substantia nigra (SN) in Parkinson’s disease (PD). Lowering PET dose reduces radiation exposure but increases noise and compromises quantification. This study evaluated a self-supervised two-step deep image prior (TS-DIP) denoising method for SV2A PET using 1/10 of the standard dose. Thirty healthy controls (HCs) and 30 PD patients underwent 60-minute PET scans to acquire full-count list-mode data, later down-sampled into ten independent 1/10-count dynamic datasets. TS-DIP was applied to denoise reduced-dose frame images, and binding potential ( BP ND ) maps were estimated. Performance was assessed by comparing group differences and correlations with motor severity against full-count results. Full-count data showed significant lower BP ND in SN (-39%, p = 0.003) and red nucleus (RN; -27%, p = 0.009) in PD versus HCs. With 1/10-count data, SN differences remained significant, but RN differences were inconsistent. TS-DIP introduced minimal bias, restored statistical significance across all noise realizations, and improved recovery of correlations with motor scores (SN: r = -0.43 ± 0.02; RN: r = -0.42 ± 0.04) compared to raw 1/10-count data. Dynamic SV2A PET imaging at substantially reduced doses is feasible when combined with advanced DL-based denoising techniques such as TS-DIP, supporting its potential for broader clinical application.
Li et al. (Sun,) studied this question.