DiGMa: A Dual-Domain Collaborative Framework for Efficient Temporal MPI Reconstruction

Magnetic particle imaging (MPI) is a novel imaging technique offering radiation-free and fast imaging with promise for temporal imaging studies (e.g., vascular and functional imaging). However, MPI temporal reconstruction is challenged by per-frame noise and the cost of joint spacetemporal modeling. In this study, we first propose a dualdomain collaborative framework for efficient temporal MPI reconstruction, named DiGMa, to address the two main challenges. DiGMa performs signal-domain denoising via GCN-based inter-frame modeling, and image-domain enhancement via Mamba that efficiently captures long-range dependencies. The proposed method was evaluated on both simulation and measured data, demonstrating superior performance over traditional approaches. Besides, qualitative results indicate fewer artifacts and superior image quality, while the computational load of the model is also reduced. DiGMa thus achieves a favorable balance between reconstruction quality and computational efficiency for temporal MPI.