Endoscopy 2022; 54(S 01): S186
DOI: 10.1055/s-0042-1745066
Abstracts | ESGE Days 2022
ESGE Days 2022 Digital poster exhibition

COMPREHENSIVE REVIEW OF PUBLICLY AVAILABLE COLONOSCOPIC IMAGING DATASETS FOR ARTIFICIAL INTELLIGENCE RESEARCH: AVAILABILITY, ACCESSIBILITY AND USABILITY

B.B. Houwen
1   Amsterdam Gastroenterology Endocrinology Metabolism, Amsterdam University Medical Centres, location Academic Medical Center, University of Amsterdam, Gastroenterology and Hepatology, Amsterdam, Netherlands
,
K.J. Nass
1   Amsterdam Gastroenterology Endocrinology Metabolism, Amsterdam University Medical Centres, location Academic Medical Center, University of Amsterdam, Gastroenterology and Hepatology, Amsterdam, Netherlands
,
J.L. Vleugels
1   Amsterdam Gastroenterology Endocrinology Metabolism, Amsterdam University Medical Centres, location Academic Medical Center, University of Amsterdam, Gastroenterology and Hepatology, Amsterdam, Netherlands
,
P. Fockens
1   Amsterdam Gastroenterology Endocrinology Metabolism, Amsterdam University Medical Centres, location Academic Medical Center, University of Amsterdam, Gastroenterology and Hepatology, Amsterdam, Netherlands
,
Y. Hazewinkel
2   Radboud University Nijmegen Medical Centre, Radboud University of Nijmegen, Gastroenterology and Hepatology, Nijmegen, Netherlands
,
E. Dekker
1   Amsterdam Gastroenterology Endocrinology Metabolism, Amsterdam University Medical Centres, location Academic Medical Center, University of Amsterdam, Gastroenterology and Hepatology, Amsterdam, Netherlands
› Author Affiliations
› Further Information
Also available at
 

Aims Publicly available datasets containing colonoscopic imaging data are valuable resources for artificial intelligence (AI)-research in gastrointestinal endoscopy. This review aimed to describe the availability, accessibility and usability of these publicly available colonoscopic imaging datasets.

Methods A systematic literature search was performed in MEDLINE and Embase to identify AI-studies describing publicly available colonoscopic imaging datasets published after 2010. Second, a targeted search using Google’s (Dataset) Search, GitHub and Figshare was done to identify datasets directly. Datasets were included if they contained data about polyp detection, polyp classification or colonoscopy quality. Datasets were categorized according to their availability as: open access, open access with barriers and regulated access. To assess the potential usability of datasets, essential details of each dataset (i.e. metadata) were extracted using a structured checklist for metadata reporting.

Results We identified 16 datasets with open access, 2 datasets open access with barriers and 12 datasets with regulated access. Thirteen open access datasets focused on polyp detection, 4 on polyp classification and 3 on colonoscopy quality (containing 14,796 images and 613 videos from≥286 patients). The proportion of metadata items reported by each of the included datasets ranged from 32% to 91%. Although technical details were in general well-reported, reporting of the annotation process and clinical information was poor.

Conclusions This review provides greater insight on the availability, accessibility and usability of colonoscopic imaging datasets as resources for AI-research. Future efforts should focus on improved reporting of metadata, maximising the potential of data resources and ultimately improving the quality of AI-research.



Publication History

Article published online:
14 April 2022

© 2022. European Society of Gastrointestinal Endoscopy. All rights reserved.

Georg Thieme Verlag KG
Rüdigerstraße 14, 70469 Stuttgart, Germany