MVKDM: A Multi-View Ki Diffusion Model for Ultra-Short Total-Body Parametric PET Imaging Using Relative Patlak Ki' Inputs.

Standard Patlak imaging requires prolonged dynamic PET acquisition ($\geq$60 min), hindering its clinical adoption. Although the relative Patlak (RP) plot shortens scan times, it entails a trade-off between acquisition duration and image fidelity, and lacks absolute quantitative validation. To address these limitations, we developed the Multi-View $K_{i}$ Diffusion Model (MVKDM), which integrates the average standardized uptake value PET (PETavg) and $K_{i}^{\prime }$ images as conditional priors to synthesize high-quality $K_{i}$ images. MVKDM employs a multi-view strategy to ensure 3D structural consistency and mitigate artifacts. The model was validated across four shortened scanning protocols (pro.1-pro.4, with durations ranging from 30 to 4 min). Assessment included visual inspection, quantitative metrics, and voxel-wise correlation/consistency analysis against reference $K_{i}$ images. MVKDM outperformed other $K_{i}$ synthesis frameworks, yielding $K_{i}$ images with clear anatomical structures and minimal noise. Across pro.1 to pro.4, MVKDM maintained high fidelity, with peak signal-to-noise ratio (PSNR) / structural similarity index (SSIM) values as follows: 45.34 $\pm$ 1.84 / 0.993 $\pm$ 0.003, 43.72 $\pm$ 2.28 / 0.990 $\pm$ 0.004, 42.78 $\pm$ 2.39 / 0.984 $\pm$ 0.006, and 42.03 $\pm$ 2.33 / 0.984 $\pm$ 0.009, respectively. Even with only 4 minutes of late-phase data, the synthetic images maintained favorable overall similarity with the reference $K_{i}$ images and preserved lesion-related uptake patterns. MVKDM enables the synthesis of high-quality and quantitatively accurate $K_{i}$ images from ultra-short dynamic scans. This framework provides a practical and robust solution for rapid total-body parametric imaging in routine clinical workflows.