一致性引导的扩散重建增强磁粒子成像中肝周肿瘤的可见性

Background: Magnetic Particle Imaging (MPI) is radiation-free and highly sensitive, enabling quantitative tracking of superparamagnetic tracers, which makes it attractive for tumor imaging. In abdominal applications, SPIO tracers accumulate in the liver via hepatic metabolism, creating dominant background that obscures nearby lesions and suppresses tumor contrast. To address this from the reconstruction side, we propose a consistency-guided diffusion method that couples a learned prior with the MPI forward model to recover concentration-faithful peri-hepatic images, without altering hardware or tracer formulation. Methods: We implement a consistency-guided diffusion procedure that alternates two steps: (i) denoising by a generative diffusion prior trained on abdominal slices, and (ii) a lightweight projection that enforces agreement with measured signals under the MPI forward operator u=Sx. A null-space stabilization term moderates trajectories when liver-tumor concentration disparity is large. Training data are synthesized from public abdominal datasets by extracting liver and tumor masks, deriving concentration maps, and forming z-axis slices; simulated signals are generated with additive white Gaussian noise. Evaluation covers (a) peri-hepatic simulation cases and (b) 3D-printed liver-tumor phantoms filled with SPIOs on an in-house MPI system. Baselines include Kaczmarz, conjugate gradient, ADMM, and a deep-equilibrium reconstruction. Results: The proposed approach improves fidelity and structure preservation versus classical solvers (higher PSNR/SSIM and lower error) and reduces the over-smoothing typical of equilibrium-based methods. In phantom studies with 5 mm and 2 mm tumor-liver separations, reconstructions exhibit suppressed background streaks/rings and clearer lesion edges, yielding consistent visual detectability across separations. Runtime overhead is modest: a single CG update per diffusion step suffices to maintain consistency. We also applied the method to in vivo human stem-cell MPI imaging experiment, yielding encouraging results and suggesting potential utility for stem cell-based cancer therapies. Conclusions: Consistency-guided diffusion reconstruction improves peri-hepatic lesion visibility in MPI under strong liver background while remaining software-only and system-agnostic. Its low operational cost and compatibility with existing scanners and tracers support translational potential for liver-adjacent oncologic imaging.