Correlative Hyperspectral Imaging Using a Dimensionality-Reduction-Based Image Fusion Method

Abstract

Chemical imaging techniques are increasingly being used in combination to achieve a greater understanding of a sample. This is especially true in the case of mass spectrometry imaging (MSI), where the use of different ionization sources allows detection of different classes of molecules across a range of spatial resolutions. There has been significant recent effort in the development of data fusion algorithms that attempt to combine the benefits of multiple techniques, such that the output provides additional information that would have not been present or obvious from the individual techniques alone. However, the majority of the data fusion methods currently in use rely on image registration to generate the fused data and therefore can suffer from artifacts caused by interpolation. Here, we present a method for data fusion that does not incorporate interpolation-based artifacts into the final fused data, applied to data acquired from multiple chemical imaging modalities. The method is evaluated using simulated data and a model polymer blend sample, before being applied to biological samples of mouse brain and lung.

Uncovering Molecular Heterogeneity in the Kidney With Spatially Targeted Mass Spectrometry

“…Examples of this have already demonstrated integration of spatially targeted MS data with autofluorescence microscopy and multiplexed immunohistochemistry approaches such as imaging mass cytometry (IMC) and co-detection by indexing (CODEX) to molecularly characterize and discover markers for kidney FTUs and cell types (Patterson et al, 2018;Singh et al, 2019;Martín-Saiz et al, 2021;Neumann et al, 2021a,b). To enable these multimodal approaches, computational tools are emerging that automatically annotate, integrate, and mine molecular imaging data from orthogonal technologies for unbiased data interpretation and identification of candidate biomarkers (Van de Plas et al, 2015;Palmer et al, 2017;Vollnhals et al, 2017;Balluff et al, 2019;Race et al, 2020;Martín-Saiz et al, 2021;Tideman et al, 2021).…”

Multimodal Imaging Based on Vibrational Spectroscopies and Mass Spectrometry Imaging Applied to Biological Tissue: A Multiscale and Multiomics Review

“…al. 275 As described earlier, with MALDI-MSI the choice of matrix partially determines what analytes are ionized and detected. Due to the lack of a "one-size-fits-all" MALDI matrix, combining data from multiple MALDI MSI acquisitions has been demonstrated by many.…”

“…This more advanced scenario will essentially mine spatial information from the modality and combine it with direct signal correlation (Figure ). Some image fusion approaches ,,, and approaches using cell morphology and intensity information , are examples of this approach. Image fusion itself is an important and growing topic (Figure ,1).…”

“…Another group qualitatively evaluated registration by marking up common ROI from two datasets achieved by the sequential digestion of tissue . An approach to minimize issues from registered data, such as the effects of image interpolation on underlying spectral data, is to forego interpolated imaging data in analysis as recently shown by Race et al…”

Correspondence-Aware Manifold Learning for Microscopic and Spatial Omics Imaging: A Novel Data Fusion Method Bringing Mass Spectrometry Imaging to a Cellular Resolution

“…Van de Plas et al have focused on the fusion of matrixassisted laser desorption ionization (MALDI) MSI data with optical images of H&E stainings through the mapping between two modalities based on linear regression models. 22 And recently, a novel method was introduced by Race et al 23 based on dimensionality reduction through non-negative matrix factorization. While providing good results, given the nonlinear nature of biological data, taking into account these complex relationships is preferable.…”