Dataset Information

HyperKvasir, a comprehensive multi-class image and video dataset for gastrointestinal endoscopy.

ABSTRACT: Artificial intelligence is currently a hot topic in medicine. However, medical data is often sparse and hard to obtain due to legal restrictions and lack of medical personnel for the cumbersome and tedious process to manually label training data. These constraints make it difficult to develop systems for automatic analysis, like detecting disease or other lesions. In this respect, this article presents HyperKvasir, the largest image and video dataset of the gastrointestinal tract available today. The data is collected during real gastro- and colonoscopy examinations at Bærum Hospital in Norway and partly labeled by experienced gastrointestinal endoscopists. The dataset contains 110,079 images and 374 videos, and represents anatomical landmarks as well as pathological and normal findings. The total number of images and video frames together is around 1 million. Initial experiments demonstrate the potential benefits of artificial intelligence-based computer-assisted diagnosis systems. The HyperKvasir dataset can play a valuable role in developing better algorithms and computer-assisted examination systems not only for gastro- and colonoscopy, but also for other fields in medicine.

SUBMITTER: Borgli H

PROVIDER: S-EPMC7455694 | biostudies-literature | 2020 Aug

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

HyperKvasir, a comprehensive multi-class image and video dataset for gastrointestinal endoscopy.

Borgli Hanna H Thambawita Vajira V Smedsrud Pia H PH Hicks Steven S Jha Debesh D Eskeland Sigrun L SL Randel Kristin Ranheim KR Pogorelov Konstantin K Lux Mathias M Nguyen Duc Tien Dang DTD Johansen Dag D Griwodz Carsten C Stensland Håkon K HK Garcia-Ceja Enrique E Schmidt Peter T PT Hammer Hugo L HL Riegler Michael A MA Halvorsen Pål P de Lange Thomas T

Scientific data 20200828 1

Artificial intelligence is currently a hot topic in medicine. However, medical data is often sparse and hard to obtain due to legal restrictions and lack of medical personnel for the cumbersome and tedious process to manually label training data. These constraints make it difficult to develop systems for automatic analysis, like detecting disease or other lesions. In this respect, this article presents HyperKvasir, the largest image and video dataset of the gastrointestinal tract available today ...[more]

PMID: 32859981

Similar Datasets

Project description:BackgroundVideo capsule endoscopy (VCE) is a commonly used test for the evaluation of obscure gastrointestinal bleeding. However, long-term outcomes of patients undergoing VCE are unclear.AimsTo evaluate the long-term outcomes in patients undergoing VCE for suspected obscure bleeding including iron deficiency anemia (IDA), and determine the need for additional intervention for persistence or recurrence of symptoms in patients with a diagnostic as well as non-diagnostic VCE.DesignRetrospective cohort study within a large county hospital system.MethodsWe collected information on indications and findings of VCE and outcomes including further testing, bleeding, and hemoglobin (Hgb) at last follow-up through structured review of the electronic health records. VCE findings were classified as active bleeding or high potential for bleeding (P2), intermediate potential (P1) or without any disruption of the mucosa, and no potential for bleeding (P0). We compared demographic and clinical characteristics between patients with and without normal Hgb at the time of last follow up.ResultsWe examined 116 patients who underwent VCEs performed for obscure gastrointestinal (GI) bleeding during 2010 to 2012 with mean duration of follow up after VCE completion of 571 days (standard deviation [SD]?=?248). Abnormal VCE findings (37.9% for P1 lesions, 44.8% for P2 lesions) were seen in 106 (87.9%) patients. Additional diagnostic testing was performed in 55/116 (47.4%) (67.7% GI procedures). Hgb was restored to normal range in 59/116 (50.9%) by end of follow up which were attributed to iron supplementation and/or discontinuation of non-steroidal anti-inflammatory drugs (NSAIDs) in a majority. Twenty six of 116 patients experienced rebleeding (22.4%).ConclusionsThe diagnostic yield of VCE is high among patients with obscure GI bleeding. More than 50% of patients achieve normal Hgb in the long term with conservative measures such as iron supplementation and the discontinuation of NSAIDs.

Project description:ImportanceEvaluation of acute gastrointestinal (GI) bleeding using invasive endoscopic procedures comprising the standard of care (SOC)-upper endoscopy and colonoscopy-can expose the endoscopy staff to SARS-CoV-2. Video capsule endoscopy (VCE) does not generate aerosols and only requires 1 person to manage the procedure.ObjectiveTo examine the safety of VCE for the initial evaluation of GI bleeding at the peak of the COVID-19 pandemic to identify signs of active bleeding while minimizing patient and personnel exposure, saving personal protective equipment, and avoiding invasive or unnecessary procedures.Design, setting, and participantsA multicenter (UMass Memorial Medical Center and Louisiana State University Health Sciences Center) retrospective cohort study including 146 patients with COVID-19 who received VCE as the first-line diagnostic modality was conducted from March 15 to June 15, 2020, compared with SOC in January 2020 for evaluation of GI bleeding. The association between treatment and outcomes was estimated using multivariable regression adjusting for potential confounders. Propensity score matching was used to verify the results.Main outcomes and measuresThe primary end point was detection of active bleeding or stigmata of recent bleeding. Secondary end points included the number of patients requiring any invasive procedures, number of additional procedures, rates of rebleeding and rehospitalization, transfusion requirements, and mortality.ResultsAmong 146 patients, 92 (63.0%) were men; mean (SD) age was 64.93 (14.13) years in the COVID-19 group and 61.33 (13.39) years in the SOC group. Active bleeding or stigmata of recent bleeding was observed in 44 (59.5%) patients in the COVID-19 group compared with 18 (25.0%) in the SOC group (adjusted odds ratio, 5.23; 95% CI, 2.23 to 12.27). Only 36 patients (48.7%) in the COVID-19 group required any invasive procedure during the hospitalization compared with 70 (97.2%) in the SOC group (adjusted odds ratio, 0.01; 95% CI, 0.001 to 0.08). The mean (SD) number of invasive procedures was 0.59 (0.77) per patient in the COVID-19 group compared with 1.18 (0.48) per patient in the SOC group (adjusted difference, -0.54; 95% CI, -0.77 to -0.31). Both approaches appeared to be safe and there was no significant difference in transfusion requirements, rebleeding, rehospitalization, or in-hospital mortality. No mortality was attributed to GI bleeding in either group.Conclusions and relevanceIn this cohort study, first-line diagnostic evaluation of acute GI bleeding using VCE appeared to be a safe and useful alternative to the traditional approach of upper endoscopy and colonoscopy. Use of VCE was associated with increased detection of active bleeding and a reduced number of invasive procedures and unnecessary exposure of personnel to SARS-CoV-2 and use of personal protective equipment.

Dataset Information

HyperKvasir, a comprehensive multi-class image and video dataset for gastrointestinal endoscopy.

Publications

HyperKvasir, a comprehensive multi-class image and video dataset for gastrointestinal endoscopy.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets