Jack4688/MSI-SuperRes-DESI

DESI-MSI Super-Resolution Dataset (200 µm ↔ 50 µm adjacent sections)

Real desorption electrospray ionization mass spectrometry imaging (DESI-MSI) data used to evaluate ×4 spatial super-resolution of MSI, as reported in our manuscript "Single-modality deep super-resolution accelerates mass spectrometry imaging" (under review). Unlike simulated down-sampling benchmarks, this dataset provides physically measured low- and high-resolution acquisitions of adjacent tissue sections, enabling a realistic test of whether a model trained on public data transfers to independently acquired hardware measurements.

What is in this dataset

For each of three organs we acquired one low-resolution (200 µm pixel) and one high-resolution (50 µm pixel) DESI-MSI image on adjacent tissue sections (10 µm thick, ≤ 30 µm apart). The 200 µm image mimics a fast, ~7–8× shorter acquisition; the 50 µm image is the high-fidelity reference. All images are provided in the open-standard imzML format.

Organ	Resolution	imzML file
Brain	200 µm	imzml/DESI-brain-02-200um.*
Brain	50 µm	imzml/DESI-brain-03-50um.*
Kidney	200 µm	imzml/DESI-kidney-01-200um.*
Kidney	50 µm	imzml/DESI-kidney-02-50um.*
Testis	200 µm	imzml/DESI-testis-20260609-01.*
Testis	50 µm	imzml/DESI-testis-20260609-02.*

Repository structure

imzml/    # open-standard imzML data; each dataset is a pair:
          #   <name>.imzML   (XML metadata)
          #   <name>.ibd     (binary spectra)
raw/      # one example Waters MassLynx .raw acquisition (200 µm brain), as a
          # format sample; readable with MassLynx / ProteoWizard MSConvert

imzML note: each .imzML metadata file must be kept together with its matching .ibd binary; both are provided in imzml/ under the same base name.

Acquisition

• Instrument: Waters Cyclic IMS (DESI source), positive-ion mode.
• Sampling: line-raster DESI; low-resolution = 200 µm probe step, high-resolution = 50 µm probe step on the adjacent section.
• Tissue: rat brain, kidney, and testis; 10 µm cryosections on glass slides.

Processing

The provided imzML files were peak-picked and TIC-normalized with standard MSI practice (Cardinal / equivalent): MAD noise estimation with SNR ≥ 3, 0.1 Da peak alignment, and retention of peaks present in ≥ 0.3 % of pixels. The pipeline that converts a Waters .raw acquisition to imzML (ProteoWizard MSConvert → raster reconstruction → peak picking → TIC normalization) is described in the manuscript and its Supporting Information, together with a validity check of the pipeline (quantitative-abundance preservation across methods).

Note on mass calibration. These acquisitions carry a systematic ~+77 ppm offset (lock-mass not applied during conversion); m/z values should be single-point recalibrated (e.g. against PC(34:1) [M+H]⁺, 760.585) before assignment.

Quick start

# pip install pyimzml numpy
from pyimzml.ImzMLParser import ImzMLParser, getionimage

p = ImzMLParser("imzml/DESI-brain-02-200um.imzML")   # .ibd is read automatically
img = getionimage(p, 760.585, tol=0.1)               # one ion image (PC(34:1) [M+H]+)

import matplotlib.pyplot as plt
plt.imshow(img, cmap="hot", origin="lower", interpolation="nearest"); plt.show()

Intended use

• Benchmarking single-modality MSI super-resolution on real hardware measurements (not simulated down-sampling).
• Cross-organ / cross-dataset generalization tests.
• The 200 µm and 50 µm images are adjacent sections, so they cannot be matched pixel-for-pixel; the paper aligns them with a single shared affine transform before computing metrics (see Supporting Information).

Related resources

• Code: https://github.com/ZZh-CSU/MSI-Super-Resolution-Reconstruction
• Model weights: Jack4688/MSI-SuperRes-DESI-weights (separate Hugging Face model repo)
• Paper: <add DOI when available>

Citation

If you use this dataset, please cite:

@article{Zhao_MSI_SR,
  title   = {Single-modality deep super-resolution accelerates mass spectrometry imaging},
  author  = {Zhao, Zhihao and others},
  journal = {Analytical Chemistry},
  year    = {2026},
  note    = {Under review}
}

License

Released under CC-BY-4.0: you may share and adapt the data provided you give appropriate credit by citing the paper above.

Contact

Zhihao Zhao — zhaozhihao946@gmail.com