This article introduces a novel experimental and computational approach for automated cell annotation and classification on standard hematoxylin and eosin (H&E) stained histopathology images, circumventing the need for error-prone human annotations. This method involves co-registering H&E images with multiplexed immunofluorescence (mIF) of the same tissue section to establish high-quality, large-scale ground truth labels for millions of cells based on lineage protein markers. A deep learning model incorporating self-supervised learning and domain adaptation is then trained to classify four major cell types—tumor cells, lymphocytes, neutrophils, and macrophages—with high accuracy. Crucially, the authors demonstrate that this single-cell spatial analysis approach can be used to discover novel spatial biomarkers in the tumor microenvironment that are linked to patient survival and prediction of response to immune checkpoint inhibitors. This scalable framework provides a foundation for single-cell analysis in precision oncology using routinely collected H&E slides.

References:

• Li Z, Mirjahanmardi S H, Sali R, et al. Automated cell annotation and classification on histopathology for spatial biomarker discovery[J]. Nature Communications, 2025, 16(1): 6240.