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

Area of Science:

• Biomedical imaging
• Computational pathology
• Mass spectrometry imaging (MSI)

Background:

• Mass spectrometry imaging (MSI) is a valuable tool for biochemical analysis but is limited by slow data acquisition.
• High-resolution tissue mapping and 3D reconstruction are hindered by the time-consuming nature of raster scanning in MSI.

Purpose of the Study:

• To develop a computational framework for reconstructing high-resolution MSI ion images from sparsely sampled data.
• To accelerate MSI data acquisition and improve the feasibility of high-resolution tissue analysis.

Main Methods:

• Integration of a raster scanning forward model with a deep learning prior (pretrained network-based denoiser).
• Utilizing a plug-and-play iterative reconstruction algorithm without the need for retraining for diverse acquisition settings.
• Testing the framework on data from various MSI instruments, acquisition parameters, and tissue types.

Main Results:

• Successful reconstruction of high-fidelity ion images from sparse MSI data across different instruments and settings.
• Demonstrated robustness of the method across biologically and structurally distinct tissues, including brain and kidney sections.
• Validation of the deep learning prior's effectiveness without additional training for new datasets.

Conclusions:

• The developed computational framework significantly enhances MSI data reconstruction from sparse sampling.
• The method offers a versatile and robust solution for high-resolution MSI, applicable across various tissues and experimental workflows.
• This approach has the potential to broaden the deployment and applications of MSI in biomedical research.