BONS-AI-TUE-AMC/GastroNet5M

📊 GastroNet-5M

A Multicenter Dataset for Developing Foundation Models in Gastrointestinal Endoscopy

Representative endoscopic images from the GastroNet-5M dataset.

---

📖 Overview

GastroNet-5M is the largest publicly available dataset of gastrointestinal endoscopic images to date.
It contains 4,820,653 unlabeled images derived from approximately 500,000 unique endoscopic procedures, collected across eight Dutch hospitals between 2012 and 2020.

The dataset covers a wide range of procedures in both the upper and lower gastrointestinal (GI) tract, acquired using endoscopy systems from all major manufacturers.

GastroNet-5M was created to accelerate the development of deep learning systems for gastrointestinal endoscopy, especially as a pretraining dataset for foundation or representation learning. It is expected to:

• Improve diagnostic accuracy
• Enhance robustness to heterogeneous imaging data
• Reduce dependence on scarce annotated datasets

---

📦 Dataset Structure

• Images are provided in PNG format.
• Stored and subdivided into zipped folders of up to 10,000 images each.
• A representative subset of 1,000 images is available for direct download.
• Each image has been anonymized to remove patient-identifying information and metadata.
• Central quality control included manual review to exclude irrelevant or sensitive images.

---

🔗 Download

The dataset can be accessed and requested via the official dataset portal:
👉 https://cortex.thetavision.nl/dataset-provider/listing/1/

The corresponding research article describing the dataset is available at:
📄 ScienceDirect: GastroNet-5M Paper

---

🧠 Applications

GastroNet-5M serves as a foundation for:

• Pretraining visual models for gastrointestinal endoscopy
• Transfer learning in downstream diagnostic tasks
• Research on data heterogeneity, robustness, and generalization

---

⚙️ Data Anonymization

All images were anonymized on-site at each hospital using proprietary software that:

• Masks patient-identifying text and metadata
• Ensures no visual identifiers remain

Some images may contain anonymization artifacts.
A central manual review was conducted to maintain dataset integrity and compliance.

---

🧾 Citation

Researchers using GastroNet-5M or related pretrained weights must cite the following papers:

Primary Dataset Paper

Jong, M. R., Boers, T. G. W., Fockens, K. N., Jukema, J. B., Kusters, C. H. J., Jaspers, T. J. M., van Eijck van Heslinga, R. A. H., Slooter, F. C., Struyvenberg, M. R., Bisschops, R., van der Putten, J. A., de With, P. H. N., van der Sommen, F., de Groof, A. J., & Bergman, J. J. (2025). GastroNet-5M: A Multicenter Dataset for Developing Foundation Models in Gastrointestinal Endoscopy. Gastroenterology. https://doi.org/10.1053/j.gastro.2025.07.030

Foundation Model Pretraining Paper

Boers, T. G. W., Fockens, K. N., van der Putten, J. A., Jaspers, T. J. M., Kusters, C. H. J., Jukema, J. B., Jong, M. R., Struyvenberg, M. R., de Groof, J., Bergman, J. J., de With, P. H. N., & van der Sommen, F. (2024). Foundation models in gastrointestinal endoscopic AI: Impact of architecture, pre-training approach and data efficiency. Medical Image Analysis, 98, 103298. https://doi.org/10.1016/j.media.2024.103298

---

📚 Recommended Related Works

Researchers are encouraged to consult these downstream studies leveraging GastroNet-5M:

1. Evaluation of an improved computer-aided detection system for Barrett's neoplasia
   Endoscopy (2025) – https://doi.org/10.1055/a-2642-7584

2. Computer-aided quality control system for Barrett's esophagus endoscopy
   Endoscopy, 57(7), 709–716 (2025) – https://doi.org/10.1055/a-2537-3510

3. Impact of standard enhancement settings of endoscopy systems on AI performance
   Endoscopy, 57(6), 602–610 (2025) – https://doi.org/10.1055/a-2530-1845

---

🪪 License & Usage

Use of GastroNet-5M is subject to the terms specified on the dataset portal.
Please review and comply with the data access agreement before downloading.