Introduction

Automatic detection of diseases by use of computers is an important, but still unexplored field of research. Such innovations may improve medical practice and refine health care systems all over the world. However, datasets containing medical images are hardly available, making reproducibility and comparison of approaches almost impossible. Here, we present Kvasir, a dataset containing images from inside the gastrointestinal (GI) tract. The collection of images are classified into three important anatomical landmarks and three clinically significant findings. In addition, it contains two categories of images related to endoscopic polyp removal. Sorting and annotation of the dataset is performed by medical doctors (ex- perienced endoscopists). In this respect, Kvasir is important for research on both single- and multi-disease computer aided detec- tion. By providing it, we invite and enable multimedia researcher into the medical domain of detection and retrieval.

Background

The human digestive system may be affected by several diseases. Altogether esophageal, stomach and colorectal cancer accounts for about 2.8 million new cases and 1.8 million deaths per year. Endoscopic examinations are the gold standards for investigation of the GI tract. Gastroscopy is an examination of the upper GI tract including esophagus, stomach and first part of small bowel, while colonoscopy covers the large bowel (colon) and rectum. Both these examinations are real-time video examinations of the inside of the GI tract by use of digital high definition endoscopes. Endoscopic examinations are resource demanding and requires both expensive technical equipment and trained personnel. For colorectal cancer prevention, endoscopic detection and removal of possible precancerous lesions are essential. Adenoma detection is therefore considered to be an important quality indicator in colorectal cancer screening. However, the ability to detect adenomas varies between doctors, and this may eventually affect the individuals’ risk of getting colorectal cancer. Endoscopic assessment of severity and sub-classification of different findings may also vary from one doctor to another. Accurate grading of diseases are important since it may influence decision-making on treatment and follow-up. For example, the degree of inflammation directly affects the choice of therapy in inflammatory bowel diseases (IBD). An objective and automated scoring system would therefore be highly welcomed. Automatic detection, recognition and assessment of pathological findings will probably contribute to reduce inequalities, improve quality and optimize use of scarce medical resources. Furthermore, since endoscopic examinations are real-time investigations, both normal and abnormal findings have to be recorded and documented within written reports. Automatic report generation would proba- bly contribute to reduce doctors’ time required for paperwork and thereby increase time to patient care. Reliable and careful docu- mentation with the use of minimal standard terminology (MST) may also contribute to improved patient follow-up and treatment. To our knowledge, a standardized and automatic reporting system that ensure high quality endoscopy reports does not exist. In order to make the health care system more scalable and cost effective, basic research in the intersection between computer science and medicine must go beyond traditional medical imaging by combining this area with multimedia data analysis and retrieval, artificial intelligence, and distributed processing. Next-generation medical big-data applications are a frontier for innovation, compe- tition and productivity, where there are currently large initiatives both in the EU and the US. In the area of multimedia research, people are starting to see the synergies between multimedia and medical systems. For automatic algorithmic detection of abnormalities in the GI tract, there have been many proposals from various research communities, especially for the topic of polyp detection. Hovever, the results are hard to reproduce due to lack of available medical data, i.e., the work listed above all use different and non-public data sets. Here, we therefore publish Kvasir: A Multi-Class Image Dataset for Computer Aided Gastrointestinal Disease Detection from the Vestre Viken Health Trust (Norway) containing not only polyps, but also two other findings, two classes related to polyp removal and three anatomical landmarks in the GI tract.

Use terms and Citations

The use of the Kvasir dataset is restricted for research and educational purposes only. The use of the Kvasir dataset for other purposes including commercial purposes is forbidden without prior written permission. In all documents and papers that use or refer to the Kvasir dataset or report experimental results based on the Kvasir dataset, a reference to the dataset paper have to be included:

Konstantin Pogorelov, Kristin Ranheim Randel, Carsten Griwodz, Sigrun Losada Eskeland, Thomas de Lange, Dag Johansen, Concetto Spampinato, Duc-Tien Dang-Nguyen, Mathias Lux, Peter Thelin Schmidt, Michael Riegler, Pål Halvorsen, Kvasir: A Multi-Class Image Dataset for Computer Aided Gastrointestinal Disease Detection, In MMSys'17 Proceedings of the 8th ACM on Multimedia Systems Conference (MMSYS), Pages 164-169 Taipei, Taiwan, June 20-23, 2017.

BibTeX:
@inproceedings{Pogorelov:2017:KMI:3083187.3083212,
author = {Pogorelov, Konstantin and Randel, Kristin Ranheim and Griwodz, Carsten and Eskeland, Sigrun Losada and de Lange, Thomas and Johansen, Dag and Spampinato, Concetto and Dang-Nguyen, Duc-Tien and Lux, Mathias and Schmidt, Peter Thelin and Riegler, Michael and Halvorsen, P{\aa}l},
title = {KVASIR: A Multi-Class Image Dataset for Computer Aided Gastrointestinal Disease Detection},
booktitle = {Proceedings of the 8th ACM on Multimedia Systems Conference},
series = {MMSys'17},
year = {2017},
isbn = {978-1-4503-5002-0},
location = {Taipei, Taiwan},
pages = {164--169},
numpages = {6},
url = {http://doi.acm.org/10.1145/3083187.3083212},
doi = {10.1145/3083187.3083212},
acmid = {3083212},
publisher = {ACM},
address = {New York, NY, USA},
}