Topological segmentation of mass spectrometry imaging data

Area of Science:

• Biomedical data analysis
• Computational biology
• Mass spectrometry imaging

Background:

• Image segmentation is crucial for mass spectrometry imaging (MSI) data processing.
• Existing methods like machine learning clustering may alter spectral data.
• Unsupervised methods are needed to maintain data integrity.

Purpose of the Study:

• To develop an unsupervised topological segmentation method for MSI data.
• To preserve the physical and chemical integrity of mass spectra.
• To enable accurate discrimination of spectrally homogeneous regions in complex tissues.

Main Methods:

• Utilized cosine similarity to identify and filter outliers and mixed-origin pixels.
• Evaluated data manifold dimensionality to define spectrally homogeneous regions.
• Applied topological segmentation to MALDI-TOF MSI data of human glial tumors.

Main Results:

• Successfully discriminated regions within aggressive human glial tumors.
• Demonstrated correlated region allocation across parallel tissue sections.
• Showcased the importance of filtering mixed cellular content for heterogeneous tumor analysis.

Conclusions:

• Topological segmentation effectively filters transition zones, yielding homogeneous data clusters.
• Identified regions represent characteristic molecular composition while preserving natural variability.
• This method is critical for identifying metabolic alterations in tumor microenvironments and metastasis.