A PET MOVIE for interpretable synthesis of late-frame [11C]-PiB PET images from early-frame counterparts

Dynamic [ 11 C]-Pittsburgh compound B (PiB) PET provides temporal information for studying amyloid-related diseases, though the long acquisition results in patient discomfort and additional operational costs. We proposed Modeling ODEs with Visually Interpretable Evolution (MOVIE), a generative framework for synthesizing late-phase [ 11 C]-PiB PET images from a short 1-min early-phase acquisition, such that multiple biomarker information can be obtained in a single session. 78 dynamic PiB PET scans from a mixed cohort of cerebral amyloid angiopathy, Alzheimer’s disease, and hypertension patients were used to obtain the early- and late-phase images. We incorporated neural ordinary or stochastic differential equations (ODEs/SDEs) in MOVIE as well as utilizing mid-phase frames to guide image prediction. The metrics PSNR, SSIM, LPIPS, effect size, and SUVR-based regional analyses were used for quantitative evaluation; classification performance for amyloid positivity was assessed using ROC AUC and decision-curve analysis (DCA). The step-by-step synthesis of images from the early- to the late-phase was visualized. MOVIE achieved quantitative gains with improved metrics and outperformed a UNet baseline model (highest PSNR: 41.46 ± 2.96 and SSIM: 0.969 ± 0.0093 for the SDE model with effect size > 0.8). Regarding classification performance, the covariate-adjusted AUC for amyloid positivity in the global Centiloid composite ROI was higher in the MOVIE (SDE) implementation (AUC = 0.881) than the baseline, and also yielded a net benefit above the baseline under DCA. Visualizations of the step-by-step generative trajectories showed biologically plausible early-to-late evolution of uptake spatial patterns. MOVIE is, to our knowledge, the first framework to couple early-to-late PET synthesis with a temporally continuous, physiology-guided generative model. By moving beyond black-box regression, MOVIE shows potential to reduce scan time, alleviate hospital operations, improve interpretability, and support clinical decision-making in amyloid PET imaging.