Changes of Mass Spectra Patterns on a Brain Tissue Section Revealed by Deep Learning with Imaging Mass Spectrometry...

Area of Science:

• Neuroscience
• Analytical Chemistry
• Computational Biology

Background:

• Tissue environments exhibit characteristic mass spectra patterns in imaging mass spectrometry (IMS).
• Visual assessment of boundaries between distinct tissue environments in IMS data remains challenging.

Purpose of the Study:

• To develop and validate a deep learning approach for visually assessing tissue boundaries using IMS data.
• To extract characteristic mass spectra patterns and segment rodent brain sections.

Main Methods:

• Application of imaging mass spectrometry (IMS) on rodent brain sections.
• Utilized convolutional neural networks (CNNs), a deep learning method, for pattern extraction and classification.
• Analyzed data from desorption electrospray ionization (DESI)-IMS and matrix-assisted laser desorption (MALDI)-IMS.

Main Results:

• CNN models achieved high accuracy and low loss rates on diverse rodent brain datasets.
• Successfully generated segmentation and classification score images based on extracted spectral features.
• Boundary imaging revealed abrupt changes in mass spectra between tissue types, indicating no significant intermediate zones.

Conclusions:

• CNN-based analysis of IMS data provides a powerful tool for visually assessing spectral pattern changes.
• This method enables precise mapping of tissue boundaries in brain sections.
• Contributes to a comprehensive understanding of tissue microenvironments and their molecular landscapes.