Digital Pathology · Aignostics
OpenTME
Pre-analysed TCGA H&E whole-slide images with tumour-microenvironment spatial features from Aignostics.
What OpenTME is
OpenTME is a digital pathology dataset published by Aignostics, designed for researchers in medicine, biology, and computational pathology. It provides pre-analysed haematoxylin and eosin (H&E) whole-slide images sourced from The Cancer Genome Atlas (TCGA).
The dataset covers multiple primary cancer types, including breast, bladder, colorectal, liver, and lung malignancies. Rather than presenting raw whole-slide images alone, it offers extracted cell-level annotations and spatial profiles of the tumour microenvironment — the complex cellular and structural network surrounding tumour tissue.
By compiling these pre-extracted features and segmentations, OpenTME supports oncology research, biomarker discovery, and spatial biology workflows without requiring researchers to generate baseline segmentations from scratch.
How it was made
Whole-slide H&E images from TCGA are processed using Atlas H&E-TME, an analytical pipeline that extracts cell-level annotations and characterises spatial relationships within the tumour microenvironment.
The resulting data are packaged as an image folder suitable for multimodal machine learning, facilitating tasks including image classification, object detection, cell and tissue segmentation, and feature extraction across whole-tumour regions and tertiary lymphoid structures.
Key facts
- Data source
- The Cancer Genome Atlas (TCGA)
- Modality
- H&E whole-slide imaging
- Cancer types
- Breast, bladder, colorectal, liver, and lung cancers
- Processing model
- Atlas H&E-TME
- Format
- imagefolder
- Supported tasks
- Image classification, segmentation, feature extraction, object detection
What you can use it for
- Tumour Microenvironment Analysis
- Investigate spatial configurations of cellular sub-populations, tertiary lymphoid structures, and whole-tumour regions across diverse cancer types.
- Histopathology Biomarker Discovery
- Correlate spatial cellular metrics and tissue architecture directly with patient oncological profiles.
- Pathology Model Development
- Train and benchmark computer vision models on standardized cell-segmentation, tissue-segmentation, and object-detection tasks.
Explore the data
This dataset is gated
Its authors require their terms to be accepted before the data can be read. Accept them on the original dataset page. The platform does not accept terms on your behalf, so the rows are not shown here.Strengths and limits
Good at
- Delivers ready-to-use cell annotations and spatial features directly derived from standard TCGA whole-slide images.
- Covers multiple major cancer types within a single standardised dataset schema.
- Integrates with standard machine-learning pipelines through imagefolder structure and ML-Croissant metadata.
Watch out for
- Extracted features and annotations are derived from the Atlas H&E-TME model rather than exhaustive manual pathologist annotations.
- Dataset scope is constrained to the cohorts, clinical profiles, and slide qualities available in TCGA.
Questions
Can I look inside OpenTME without downloading it?
Not here: its authors gate it behind terms each user must accept on the original page. Its description and statistics are still shown.
What licence is OpenTME under?
Its authors declare other. Read the terms on the original page before reusing the data.
Which cancer types are covered in OpenTME?
The dataset focuses on breast, bladder, colorectal, liver, and lung cancers from TCGA cohorts, with tags indicating relevance across broader oncology indications.
What features are provided alongside the images?
OpenTME includes cell-level annotations, tissue segmentations, tertiary lymphoid structure indicators, and spatial tumour-microenvironment features generated by Atlas H&E-TME.
What machine learning tasks is the dataset suited for?
It is configured for image classification, cell and tissue segmentation, object detection, and feature extraction tasks.
How to cite it
arXiv:2604.12075
Written from OpenTME's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.