Pseudo-spectral angle mapping for pixel and cell classification in highly multiplexed immunofluorescence images.

Area of Science:

• Computational pathology
• Spatial biology
• Biomedical imaging analysis

Background:

• Highly multiplexed microscopy generates rich spatial data for studying cells in native tissues.
• Analyzing high-content images from these techniques requires robust and generalizable computational tools.
• Existing methods struggle to evaluate cellular constituents and stroma in high-plex imaging data.

Purpose of the Study:

• To adapt spectral angle mapping for high-plex immunofluorescence (IF) image analysis.
• To develop a robust and flexible method for pixel classification in high-plex images.
• To enable accurate classification of individual cells within complex tissue environments.

Main Methods:

• Adapted spectral angle mapping (SAM) into pseudo-spectral angle mapping (pSAM) to compress the channel dimension of high-plex IF images.
• Generated pixel classifications using pSAM class maps.
• Combined pSAM-derived class maps with instance segmentation (Cellpose2.0) for cell classification.

Main Results:

• In colon biopsies (13-plex), pSAM generated 16 class maps, enabling cell classification into 13 cell types with Cellpose2.0.
• In kidney biopsies (44-plex), pSAM plus Cellpose2.0 detected 38 diverse structural and immune cell classes.
• Achieved high performance scores (0.85 and 0.82) in both colon and kidney datasets, demonstrating robustness.

Conclusions:

• pSAM is a powerful and generalizable tool for analyzing high-plex IF image data.
• pSAM facilitates the classification of cells in high-dimensional images.
• This method enhances the study of cellular and tissue architecture in spatial biology.