Machine learning-based identification of H&E-derived morphologic features associated with CD8+ T cell immune exclusion

Immune exclusion, characterized by a predominance of CD8+ T cells in the stroma of the tumor microenvironment (TME) that are not in contact with tumor cells, is associated with a lack of patient response to immune checkpoint inhibitor therapy.

Artificial intelligence (AI) models have shown promise in quantifying tissue and cellular features from images of hematoxylin and eosin (H&E)-stained slides without the need for immunostaining, but they fall short in identifying specific cell subtypes (e.g., CD8+ T cells vs. other lymphocytes).

Here, we describe the development of a machine learning model that can classify CD8-based immune phenotypes (IP) using features extracted from H&E images alone.