Graph-based description of tertiary lymphoid organs at single-cell level.

Area of Science:

• Computational pathology
• Immunohistochemistry
• Bioinformatics

Background:

• Observer-dependent immune cell evaluation in microscopy is subjective.
• Analyzing spatial patterns of lymphocytic infiltrates is crucial for therapeutic decisions in oncology and transplantation.
• Reproducible quantitative measures are needed to complement existing methods.

Purpose of the Study:

• To develop a reproducible, graph-based method for assessing the spatial composition of lymphocytic infiltrates in whole slide images.
• To quantify and classify immune cell clustering patterns for improved diagnostic accuracy.
• To address the need for scalable and adjustable frameworks in computational pathology.

Main Methods:

• Developed a graph-based approach using Delaunay triangulation and distance criteria on cell coordinates from whole slide images.
• Utilized neighborhood graphs, node/edge composition, and support vector machines for classification.
• Incorporated automated cell detection, region identification, and classification of lymphocytic clusters by organization degree.

Main Results:

• Demonstrated the method's ability to describe variability in lymphocytic infiltrates across different tissues (renal allografts, breast cancer, cystic fibrosis lung).
• Introduced a neighborhood feature to distinguish phenotypically different immune infiltrates based on graph edge types.
• Successfully classified lymphocytic clusters according to their degree of organization.

Conclusions:

• The graph-based assessment provides reproducible measures for spatial immune cell composition, complementing subjective methods.
• The scalable framework can be adapted for various research questions in immunology and pathology.
• This approach offers a significant advancement for analyzing immune infiltrates in clinical and research settings.