# Image Fusion for Super‐Resolution Mass Spectrometry Imaging of Plant Tissue

## TL;DR

This paper introduces a new deep learning method to enhance the resolution of plant mass spectrometry imaging by combining chemical and morphological data.

## Contribution

The novel edge perceptual loss metric and LCRN workflow enable high-quality super-resolution fusion of plant MSI data.

## Key findings

- LCRN achieves up to 20-fold magnification in plant MSI super-resolution.
- The edge perceptual loss metric preserves complex plant tissue morphology better than standard metrics.
- Fusion results effectively combine chemical and morphological information from MSI and microscopy.

## Abstract

Mass spectrometry imaging (MSI) is a vital tool in botanical research. Image fusion is introduced for resolution enhancement of MSI data from animal samples, but its application to plant MSI data resulted in unsatisfactory visualizations due to the distinct morphological characteristics of plant tissues. Herein, this study presents loss controlled residual network (LCRN), a workflow dedicated to the super‐resolution fusion of plant MSI data. The pipeline used a residual connection‐based neural network implemented with a novel loss metric called edge perceptual loss. Edge perceptual loss is developed for evaluating complex morphological information that can not be properly reflected by common image metrics, and its implementation in loss propagation is vital to the quality of the fusion result. Compared to existing deep learning‐based methods, LCRN is able to generate a high‐quality super‐resolution fusion image of extra high magnification (up to 20‐fold) that combined chemical and morphological information obtained from MSI and microscopy, respectively.

A loss controlled residual network (LCRN) workflow is developed for super‐resolution fusion of plant mass spectrometry imaging data. LCRN uses a novel edge perceptual loss metric to preserve complex plant tissue morphology. LCRN achieves up to 20‐fold magnification while effectively combining chemical information from mass spectrometry with morphological details from microscopic images.