Reliable automated identification of tumor-associated macrophages in H&E-stained DLBCL sections

Develop reliable automated methods for identifying and quantifying M2-polarized tumor-associated macrophages directly in hematoxylin-and-eosin-stained diffuse large B-cell lymphoma tissue sections despite morphological overlap with other myeloid and lymphoid cell types and the scarcity of expert-annotated datasets.

Background

The paper identifies automated identification of M2-polarized tumor-associated macrophages in H&E-stained diffuse large B-cell lymphoma sections as unresolved because macrophages lack a unique or stable H&E morphology. Their appearance overlaps with other myeloid subsets and reactive lymphocytes, while existing datasets such as DLBCL-Morph and LyNSeC lack labels specific to the immunosuppressive M2 tumor-associated macrophage population.

This unresolved problem motivates the comparative evaluation of U-Net, Swin-U-Net, Cerberus-U-Net3+, YOLOv11, and HoVer-Net for direct H&E-based macrophage segmentation and quantification, with the goal of developing a scalable surrogate for CD163 immunohistochemistry.