GastroVision: A Multi-class Endoscopy Image Dataset for Computer Aided Gastrointestinal Disease Detection ### [[Preprint]](https://arxiv.org/abs/2307.08140) --------- We present an open-access multi-class GI endoscopy dataset, namely, Gastrovision, containing 8,000 images with 27 classes from two hospitals in Norway and Sweden. The dataset exhibits a diverse range of classes, including anatomical landmarks, pathological findings, polyp removal cases and normal or regular findings, and covers a wider range of clinical scenarios encountered from the endoscopic procedures. Alternatively, the dataset is also available at https://drive.google.com/drive/folders/1T35gqO7jIKNxC-gVA2YVOMdsL7PSqeAa?usp=sharing ![Classes of GI tracts provided in the dataset. A general division of different findings in the upper and lower GI tract is presented.][1] Figure 1: Classes of GI tracts provided in the dataset. A general division of different findings in the upper and lower GI tract is presented. We evaluated a series of deep learning baseline models on standard evaluation metrics using our proposed dataset. With this baseline, we invite the research community to improve our results and develop novel GI endoscopy solutions on our comprehensive set of GI finding classes. Additionally, we encourage computer vision and machine learning researchers to validate their methods on our open-access data for a fair comparison. This can aid in developing state-of-the-art solutions and computer-aided systems for GI disease detection and other general machine learning classification tasks. Please download the dataset below with a single click. Dataset ----------- * **gastrovision.zip**: contains 8000 endoscopic images. * ![enter image description here][2] Figure 2: The figure shows the number of images per class, excluding the normal class. Some classes have few samples, as it is challenging to obtain such samples in an endoscopic procedure. More information on dataset ---------------------------- Dataset resolution Total images: 8000 Broken images: 0 Unique resolutions: 7 Class: Accessory tools Total images: 1266, Resolutions in this class: {(768, 576), (720, 576), (1350, 1080), (1920, 1072), (1280, 1024)} Class: Angioectasia Total images: 17, Resolutions in this class: {(1350, 1080), (1350, 1064), (768, 576), (720, 576)} Class: Barrett's esophagus Total images: 95, Resolutions in this class: {(768, 576), (1280, 1024), (720, 576), (1920, 1072)} Class: Blood in lumen Total images: 171, Resolutions in this class: {(768, 576), (720, 576), (1350, 1080), (1350, 1064), (1920, 1072), (1280, 1024)} Class: Cecum Total images: 113, Resolutions in this class: {(1232, 1048), (768, 576), (720, 576), (1350, 1080), (1350, 1064), (1920, 1072)} Class: Colon diverticula Total images: 29, Resolutions in this class: {(1232, 1048), (768, 576), (720, 576), (1350, 1080), (1350, 1064)} Class: Colon polyps Total images: 817, Resolutions in this class: {(1232, 1048), (768, 576), (720, 576), (1350, 1080), (1350, 1064), (1920, 1072), (1280, 1024)} Class: Colorectal cancer Total images: 146, Resolutions in this class: {(768, 576), (720, 576), (1350, 1080), (1350, 1064), (1920, 1072), (1280, 1024)} Class: Duodenal bulb Total images: 205, Resolutions in this class: {(768, 576), (1280, 1024), (720, 576), (1920, 1072)} Class: Dyed-lifted-polyps Total images: 141, Resolutions in this class: {(1350, 1080), (768, 576), (720, 576)} Class: Dyed-resection-margins Total images: 245 Resolutions in this class: {(1350, 1080), (768, 576), (720, 576), (1920, 1072)} Class: Erythema Total images: 15, Resolutions in this class: {(1350, 1064), (768, 576), (1350, 1080), (1232, 1048)} Class: Esophageal varices Total images: 7, Resolutions in this class: {(768, 576), (1280, 1024)} Class: Esophagitis Total images: 107, Resolutions in this class: {(768, 576), (1280, 1024), (720, 576), (1920, 1072)} Class: Gastric polyps Total images: 65, Resolutions in this class: {(768, 576), (720, 576), (1350, 1064), (1920, 1072), (1280, 1024)} Class: Gastroesophageal_junction_normal z-line Total images: 330, Resolutions in this class: {(768, 576), (1280, 1024), (720, 576), (1920, 1072)} Class: Ileocecal valve Total images: 200, Resolutions in this class: {(1232, 1048), (768, 576), (720, 576), (1350, 1080), (1350, 1064), (1920, 1072)} Class: Mucosal inflammation large bowel Total images: 29, Resolutions in this class: {(1350, 1064), (768, 576), (1350, 1080), (720, 576)} Class: Normal colon Total images: 326, Resolutions in this class: {(768, 576), (720, 576), (1350, 1080), (1350, 1064), (1920, 1072), (1280, 1024)} Class: Normal esophagus Total images: 140, Resolutions in this class: {(768, 576), (1280, 1024), (720, 576), (1920, 1072)} Class: Normal mucosa and vasular pattern in the large bowel Total images: 1141, Resolutions in this class: {(768, 576), (720, 576), (1350, 1080), (1350, 1064), (1920, 1072), (1280, 1024)} Class: Normal stomach Total images: 969, Resolutions in this class: {(768, 576), (1280, 1024), (720, 576), (1920, 1072)} Class: Pylorus Total images: 393, Resolutions in this class: {(768, 576), (1280, 1024), (720, 576), (1920, 1072)} Class: Resected polyps Total images: 90, Resolutions in this class: {(768, 576), (1280, 1024), (720, 576), (1920, 1072)} Class: Resection margins Total images: 24, Resolutions in this class: {(768, 576), (720, 576)} Class: Retroflex rectum Total images: 67, Resolutions in this class: {(1350, 1080), (768, 576), (720, 576), (1920, 1072)} Class: Small bowel_terminal ileum Total images: 846, Resolutions in this class: {(768, 576), (1280, 1024), (720, 576), (1920, 1072)} Class: Ulcer Total images: 6, Resolutions in this class: {(768, 576), (1280, 1024)} ![enter image description here][3] Figure 3: Resolutions of the 8,000 images of GastroVision. Terms of use --------- C-By Attribution 4.0 International. The use of the dataset is restricted to research and education purposes. The use of the dataset is forbidden for commercial use without prior written permission. For other purposes, contact us (see below). In all documents and publications that use the dataset or report experimental results based on the dataset, a reference to the dataset paper has to be included. Please email debesh.jha@northwestern.edu if you have any questions regarding how to cite the dataset. For commercial questions please mail paalh@simula.no Ethics approval In this study, we used fully anonymized data approved by Privacy Data Protection Authority. It was exempted from approval from the Regional Committee for Medical and Health Research Ethics - South East Norway. Furthermore, we confirm that all experiments were performed in accordance with the relevant guidelines and regulations of the Regional Committee for Medical and Health Research Ethics - South East Norway, and the GDPR. Citation: --------- @proceedings{jha2023gastrovision, title={GastroVision: A Multi-class Endoscopy Image Dataset for Computer Aided Gastrointestinal Disease Detection}, author={Jha, Debesh and Sharma, Vanshali and Dasu, Neethi and Tomar, Nikhil Kumar and Hicks, Steven and Bhuyan, MK and Das, Pradip K and Riegler, Michael A and Halvorsen, P{\aa}l and de Lange, Thomas and others}, booktitle={ICML Workshop on Machine Learning for Multimodal Healthcare Data (ML4MHD 2023)}, year={2023} } Contact --------- Email debesh.jha@northwestern.edu if you have any questions about the dataset and our research activities. paalh@simula.no should be contacted in case of commercial interests with the data. We always welcome collaboration and joint research! [1]: https://files.osf.io/v1/resources/84e7f/providers/osfstorage/6502d2add9f2c96787d04b72?mode=render [2]: https://files.osf.io/v1/resources/84e7f/providers/osfstorage/6502d4a6d9f2c967a0d04b13?mode=render [3]: https://files.osf.io/v1/resources/84e7f/providers/osfstorage/6502d40c767f4a30efde8f2f?mode=render