# Evaluating Batch Correction Methods for Large-Scale Mass Spectrometry Imaging of Heterogeneous Tissues

## TL;DR

This paper evaluates methods to correct batch effects in mass spectrometry imaging data to improve its reliability for large-scale studies.

## Contribution

The work introduces a pixel-by-pixel evaluation framework for batch correction in mass spectrometry imaging.

## Key findings

- Batch correction methods were assessed for their ability to stabilize intensity variability in MSI data.
- The study highlights the unique challenges of applying batch correction to spatially resolved MSI data.
- A systematic approach is proposed to evaluate correction methods while preserving spatial variability.

## Abstract

In mass spectrometry
imaging (MSI), the fluctuation in detected
ion intensities, which is associated with “technical factors”
and not the variability of molecular composition of the sample itself,
may be referred to as “batch effects”. These batch effects
are a major barrier to the more widespread uptake and use of MSI for
larger clinical and preclinical studies. In other fields, such as
metabolomics and transcriptomics, batch correction methods have been
introduced and commonly adopted. These methods aim to mitigate systematic
biases introduced by differences in experimental conditions, instruments,
or processing batches in high-dimensional data, such as omics or imaging
data sets. Mass spectrometry imaging poses additional challenges compared
to these fields such as the need to ensure that expected intensity
fluctuations throughout a sample, associated with expected spatial
variability, are maintained and the inability to randomly introduce
quality control spectra. To date, there is no widely adopted approach
to the batch correction of mass spectrometry imaging data. In this
work, we consider both stabilization of intensity variability and
the usefulness of correction methods for spatially resolved data.
We present a pixel-by-pixel evaluation of batch correction for mass
spectrometry imaging data.