High-resolution panoramic segmentation map of the tumor microenvironment

The analysis of tumor microenvironment is an important topic in modern digital pathology, but there is currently no publicly available high-resolution histological atlas. We established a semantic segmentation model involving 15 common components of pathological slices, and applied it to breast cancer, lung adenocarcinoma, colon adenocarcinoma, stomach adenocarcinoma, and pancreatic cancer of TCGA, and obtained a panoramic segmentation map at 0.5um/px high resolution. The categories involved in the atlas include epithelioid tissue, inflammatory regions, necrosis, fat, fibrous stroma, muscle, nerves, etc. Based on this atlas, key indicators such as tumor stromal ratio and tumor infiltrating lymphocytes can be directly quantified, providing highly interpretable tumor microenvironment information for prognosis prediction and disease intervention. We will make these maps public as fundamental resources in the field of computational pathology to facilitate in-depth analysis of the tumor microenvironment and the development of more advanced models.