Integrating Model-Based Reconstruction and Deep Learning for Accelerating Mass Spectrometry Imaging

Mass spectrometry imaging (MSI) is a powerful multiplexed biochemical imaging modality. It relies on raster scanning for localized data acquisition, which can be time-consuming, limiting applications of high-resolution tissue mapping and 3D reconstruction. This work presents a computational framework that integrates a raster scanning forward model with a deep learning prior to reconstruct high-resolution ion images from sparsely sampled pixels. The deep learning prior, implemented as a pretrained network-based denoiser, is incorporated into a plug-and-play-based iterative reconstruction algorithm without retraining for different acquisition settings. We show that our method can reconstruct high-fidelity ion images from sparse data acquired with different MSI instruments, acquisition settings, and tissue types without requiring additional training. Notably, our approach generalizes robustly to biologically and structurally distinct tissues, such as from brain to kidney sections, highlighting its potential for broad deployment in various experimental MSI workflows.