MSI-Segmentation: A Web-based MicroApp for Automated Exploration and Material Segmentation of MS Imaging Data

Significance of the Topic

Mass spectrometry imaging (MSI) offers an unprecedented ability to map the spatial distribution of hundreds to thousands of molecular species directly from biological and material surfaces. By revealing localized chemical heterogeneity, MSI drives discoveries in fields ranging from disease pathology to materials characterization. However, the sheer volume and complexity of MSI datasets demand advanced computational tools to uncover meaningful patterns without extensive manual intervention.

Study Objectives and Overview

This work introduces MSI-Segmentation, a web-based micro-application designed to streamline the exploratory analysis of preprocessed MSI data. The primary aims are:

• To automate pixel-level classification of chemically distinct regions via unsupervised learning.
• To cluster co-localized ion images for efficient identification of related molecular patterns.
• To provide easy export of segmentation and clustering outputs for downstream analyses.

Methodology

The application workflow consists of the following steps:

• Data ingestion from ASCII text files containing x, y coordinates and m/z intensities.
• Dimensionality reduction using UMAP to project high-dimensional spectra into a lower-dimensional embedding.
• Clustering of pixel spectra with HDBSCAN to detect spatially coherent chemical regions.
• Hierarchical (Ward’s) clustering of ion image distributions to group analytes sharing similar spatial patterns.
• Generation of correlation maps to compute R² values between any selected pixel spectrum and all others.
• Export of cluster masks, average spectra, and analyte masks in CSV format for visualization or further processing.

Used Instrumentation

The MSI datasets in this study were acquired under the following conditions:

• Ion source: DESI (desorption electrospray ionization) XS.
• Mass spectrometer: SELECT SERIES MRT.
• Ionization mode: negative.
• Cone voltage: 40 V; capillary voltage: 0.8 kV.
• Nebulizing gas: nitrogen at 0.8 bar.
• Spatial resolution: 30 μm pixel size; scan rate: 2 pixels per second.

Main Results and Discussion

MSI-Segmentation features three interactive modules:

• Tissue Segmentation: Unsupervised classification of pixels into discrete chemical clusters, with exportable average spectra and binary masks for overlay in HDI software.
• Analyte Clustering: Grouping of ion images by spatial distribution, generation of average ion maps, exportable peak lists per cluster, and optional thresholding to create analyte masks. Correlation analysis assists in evaluating cluster coherence.
• Spectral Correlation Map: On-the-fly computation of correlation coefficients between a chosen target pixel spectrum and all other spectra, facilitating targeted exploration of spatially correlated molecules.

Benefits and Practical Applications

MSI-Segmentation democratizes advanced MSI data analysis by offering:

• An intuitive web interface eliminating the need for multiple specialized tools.
• Automated, scalable workflows suitable for both novice and expert users.
• Versatile outputs that integrate seamlessly into broader pipelines for quantitative or targeted follow-up studies.

Future Trends and Opportunities

Potential developments include:

• Integration with 3D MSI datasets to enable volumetric segmentation.
• Incorporation of deep learning models for improved feature extraction and annotation.
• Real-time analytics via direct linkage with instrument control software.
• Expansion of compatibility to additional MSI platforms and file formats.

Conclusion

MSI-Segmentation offers a robust, web-based solution for the rapid and unsupervised exploration of MSI data. By combining UMAP, HDBSCAN, hierarchical clustering, and correlation mapping, this micro-app empowers researchers to identify chemically distinct regions and co-localized analytes efficiently, thereby accelerating insights in both biological and materials research.