Automated macrophage counting in DLBCL tissue samples: a ROF filter based approach

Area of Science:

• Computational pathology
• Image analysis
• Tumor microenvironment research

Background:

• Accurate quantification of macrophage subtypes in diffuse large B-cell lymphoma (DLBCL) tissue is crucial for understanding the tumor microenvironment.
• Current methods like gene expression analysis provide indirect information, while manual counting is labor-intensive and fails to capture heterogeneity.
• Existing commercial software for analyzing immunohistochemically stained tissue has limitations in handling variability in cell shape and staining intensity.

Purpose of the Study:

• To develop and validate an automated segmentation approach for analyzing macrophage subtypes in DLBCL tissue samples.
• To compare the performance of the novel method against manual counts and commercial software solutions.
• To assess the correlation of image morphometry data with gene expression data for a comprehensive understanding of macrophage distribution.

Main Methods:

• A Rudin-Osher-Fatemi (ROF) filter-based segmentation approach was developed, incorporating floating intensity thresholding and rule-based feature detection.
• The method was validated against manual counts and compared with two commercial software kits (Tissue Studio 64, Halo) and a machine-learning approach using 50 test images.
• The ROF method and commercial packages were applied to 44 whole tissue sections, with outputs compared to gene expression data; subsampling strategies were also tested on expert-specified tumor subregions.

Main Results:

• The novel ROF-based approach demonstrated the highest correlation with manual counts (0.9297) among all tested methods.
• Automated detection of evaluation subregions was found to be fully reliable, with subsampling yielding results nearly identical to full sampling.
• Comparison with gene expression data showed moderate to low correlation, suggesting image morphometry provides independent information on macrophage distribution.

Conclusions:

• The ROF-based approach is a successful, fully automated method for detecting IHC-stained macrophages in DLBCL tissue, competing effectively with commercial software.
• This method is externally repeatable, independent of training data, and fully documented, offering significant advantages over existing tools.
• Image morphometry, as demonstrated by the ROF method, offers an independent and valuable source of information regarding macrophage polarization and distribution within the tumor microenvironment.