cs.CVApr 13, 2026
SaveOpenTME: An Open Dataset of AI-powered H&E Tumor Microenvironment Profiles from TCGA
Organizations: Aignostics, Germany · Institute of Pathology, Charité – Universitätsmedizin Berlin, Germany · Department of Laboratory Medicine and Pathology, Mayo Clinic, Rochester, MN, US · Machine Learning Group, Technische Universität Berlin, Germany ·
Abstract
The tumor microenvironment (TME) plays a central role in cancer progression, treatment response, and patient outcomes, yet large-scale, consistent, and quantitative TME characterization from routine hematoxylin and eosin (H&E)-stained histopathology remains scarce. We introduce OpenTME, an open-access dataset of pre-computed TME profiles derived from 3,634 H&E-stained whole-slide images across five cancer types (bladder, breast, colorectal, liver, and lung cancer) from The Cancer Genome Atlas (TCGA). All outputs were generated using Atlas H&E-TME, an AI-powered application built on the Atlas family of pathology foundation models, which performs tissue quality control, tissue segmentation, cell detection and classification, and spatial neighborhood analysis, yielding over 4,500 quantitative readouts per slide at cell-level resolution. OpenTME is available for non-commercial academic research on Hugging Face. We will continue to expand OpenTME over time and anticipate it will serve as a resource for biomarker discovery, spatial biology research, and the development of computational methods for TME analysis.
Figures & tables
Figure 1: Two TCGA examples from the OpenTME dataset with Atlas H&E-TME Tissue QC, Tissue Segmentation, and Cell Classification model outputs. Model outputs are integrated to derive over 4,500 quantitative readouts per slide at cell-level resolution, including spatial neighborhood features.
| Cancer type | TCGA project(s) | Cases | Slides | Slides (final) |
| Bladder | TCGA-BLCA | 386 | 457 | 457 |
| Breast | TCGA-BRCA | 1062 | 1133 | 1125 |
| Colorectal | TCGA-COAD | 451 | 459 | 442 |
| TCGA-READ | 165 | 166 | 158 | |
| Liver | TCGA-LIHC | 365 | 379 | 372 |
| TCGA-CHOL | 39 | 39 | 39 |
Table 1: TCGA slides included in OpenTME.
Appendix figures & tables5 assets
Supplementary material from the paper’s appendix.
Appendix
| Region | Description |
| Valid Tissue | Well-preserved and clearly identifiable tissue suitable for downstream analysis. In the subsequent steps only regions identified as Valid Tissue are processed. |
| Out-of-Focus | Blurry or unclear region due to imaging issues. |
| Tissue Artifact | Distorted or altered tissue caused by staining, processing, or technical errors. |
| Marker | Physical drawings that a pathologist made on the glass slide with a pen. |
Table 2: Description of Tissue Quality Control model classes.
| Tissue Type | Description |
| Carcinoma | Regions containing carcinoma and immune cells. In case of squamous cell carcinoma this category also includes intra- and extracellular keratin stemming from squamous carcinoma; dysplastic serous epithelium featuring high-grade architectural and/or cytological atypia but without proven invasive growth. |
| Epithelial Tissue | Regions containing epithelial and immune cells. Includes: mesothelial lining of peritoneal surfaces; epithelial cell layer with ciliated/squamous/epithelium and enclosed immune cells; serous glandular cells including lumen, inner epithelial layer, outer basal layer and enclosed immune cells; mucinous glandular cells other than Brunner glands including lumen, inner epithelial layer, outer basal layer and enclosed immune cells; ductal epithelium including lumen, inner epithelial layer, outer basal layer and enclosed immune cells. |
| Stroma | Regions containing connective tissue and immune cells. Also includes secondary and tertiary lymphoid structures. |
| Necrosis | Regions of dying or dead cells. Includes coagulative, liquefactive, caseous, or fat necrosis with no discernible cell nuclei as well as partly discernible, potentially fragmented nuclei and enclosed immune cells. |
| Blood | Regions with accumulation of erythrocytes and/or fibrin, as seen in vessel lumina or hemorrhage. |
| Vessel | All vascular structures, including arteries, veins, arterioles, venules, capillary-like vessels ( 10 µm, 5 erythrocytes, single endothelial layer, no tunica media/adventitia), and lymphatic vessels. Includes arteries with round lumens, thick muscular walls, and visible elastica; veins with irregular lumens, thinner muscular walls, and no visible elastica; lymphatics with irregular lumens, thin walls, and lymphatic fluid. |
Table 3: Description of Tissue Segmentation model classes.
| Cell Type | Description |
| Carcinoma cell | Malignant epithelial cell with uncontrolled growth and potential for invasion. Includes: carcinoma with glandular differentiation; carcinoma with squamous differentiation; carcinoma with squamous differentiation and intra- or extracellular keratin accumulation; non-small cell tumor with neuroendocrine differentiation; signet-ring cell carcinoma with cytoplasmic mucin and an eccentric, compressed nucleus; mucinous adenocarcinoma; carcinoma with cytoplasmic clearing; atypical serous cells with high-grade dysplasia without proven invasive growth; and atypical squamous cells with high-grade dysplasia without proven invasive growth, with possible keratinization. |
| Epithelial cell | Structural cells forming protective layers on organs and internal surfaces as well as comprising parenchyma of most organs. Includes: nonspecific benign epithelial cells; flat squamous-type cells, with or without intra-/extracellular keratin; columnar polarized cells with hair-like projections; mucin-containing polarized cells; basal cuboidal cells at the basement membrane; glandular pyramid-shaped cells with basally placed nuclei and zymogen granules; glandular cells with oval nuclei and mucin granules; small hormone-releasing cells with round/oval nuclei and salt-and-pepper chromatin; thin spindle-shaped cells adjacent to glandular epithelium; variably shaped duct-lining cells; and elongated flattened cells lining pleural and peritoneal surfaces. |
| Fibroblast | Connective tissue cell responsible for collagen production and extracellular matrix maintenance. |
| Lymphocyte | Small immune cells involved in adaptive immunity, including T and B cell responses. Includes: immune cells with small round nuclei and inconspicuous cytoplasm and activated lymphocytes (centroblasts and centrocytes) within the germinal center with large, cleaved or vesicular nuclei, nucleoli, and distinguishable cytoplasm. |
| Plasma cell | Differentiated B cell specialized in antibody production for adaptive immune defense. |
| Macrophage | Large immune cell of innate immune response that engulfs pathogens and debris while regulating immune responses. |
Table 4: Description of Cell Classification model classes.
| Column Name | Description |
|---|---|
| TCGA_FILE_NAME | File name of the slide according to TCGA (GDC Portal). |
| TCGA_SLIDE_UUID | Slide UUID according to TCGA (GDC Portal). |
| TCGA_CASE_ID | Case ID according to TCGA (GDC Portal). |
| TCGA_PROJECT_ID | Project ID according to TCGA (GDC Portal). |
| INDICATION | Cancer type on the slide. |
| IMAGE_RESOLUTION | Microns per pixel (MPP) value of the slide. |
Table 5: Slide-level tissue and cell readouts.
| Column Name | Description |
|---|---|
| The following columns represent statistics of the neighborhood of a cell type A. The columns are | |
| present for each tissue type, cell type, and distance threshold (20 µm and 40 µm). Columns are | |
| postfixed with AROUND<cell-type-A>IN<tissue-type>_<distance-threshold> . | |
| AREA | Area around cell type A cells within <threshold> µm, contained within the <tissue-type> in µm 2 . Example: AREA_AROUND_GRANULOCYTE_IN_CARCINOMA_20 . |
| TOTAL_NUMBER_OF_NEIGHBORS | Total number of cells next to type A cells within <threshold> µm, contained within the <tissue-type> . Example: TOTAL_NUMBER_OF_NEIGHBORS_AROUND_PLASMA_CELL_IN_CARCINOMA_20 . |
| The following columns represent the relationship of a cell type A and another cell type B. | |
Table 6: Neighborhood readouts.
More in this topic
Computational Pathology
Computational pathology uses algorithms to analyze digitized tissue images for diagnosis, prognosis, biomarker discovery, and clinical decision support.
See all 279 papers