Patients participating to this study will provide images and videos of capsule endoscopy to train, tune and evaluate technological bricks of artificial intelligence solutions, in order to improve diagnostic performances of the procedure, while reducing reading time by physicians.
Capsule endoscopy is a minimally-invasive diagnostic procedure based on the ingestion (or endoscopic delivery) of a miniaturized biocompatible, camera. Capsules capture tenths of thousands images of the digestive tract. Reading the captured images and reporting is long, tedious, and at risk of errors when the reader's attention is disturbed. Artificial intelligence is expected to alleviate these limitations, by both improving diagnostic performances of capsule endoscopy while reducing reading time. Any patient in whom a capsule endoscopy examination is performed as part of routine care will be invited to participate to the study. Their de-identified images and videos will be extracted, thus allowing the creation of several databases for training, tuning and testing technological bricks of artificial intelligence. Basic clinical data will be collected (age, gender, indication for capsule endoscopy, type of device, ingestion or delivery of capsule). Images and videos will be characterized centrally and consensually by a panel of 3 expert readers, according to their level of relevance in relation to the type and indication of capsule endoscopy. The various, developed technological bricks will aim to automatically detect and characterize anatomical landmarks and abnormal findings, and to quote the intestine cleanliness. Assessment of diagnostic performance and reading time will be performed within a few months or up to five years, for each technological brick, individually and then combined, according to their stepwise development.
Study Type
OBSERVATIONAL
Enrollment
10,000
Centre d'Endoscopie Digestive Hôpital Saint-Antoine
Paris, France
RECRUITINGDevelop an artificial intelligence solution to help with diagnosis in VCE
Sensitivity in detection of lesions of intermediate to high relevance as identified centrally and consensually by a panel of 3 expert readers (AI vs standard reading)
Time frame: Through study completion, 5 years
Evaluation of the diagnostic performance of the A.I. solution in terms of detection
Other diagnostic performance (specificity, positive and negative predictive values, accuracy) in detection of lesions of intermediate to high relevance as identified centrally and consensually by a panel of 3 expert readers (AI vs standard reading)
Time frame: Through study completion, 5 years
Evaluation of the diagnostic performance of the A.I. solution in terms of characterisation
Time needed for detection of lesions of intermediate to high relevance as identified centrally and consensually by a panel of 3 expert readers (AI vs standard reading)
Time frame: Through study completion, 5 years
Evaluation of the diagnostic performance of the A.I. solution in terms of recognition of anatomical landmarks
Diagnostic performances in positioning anatomical landmarks (1st gastric, small bowl and colonic images) as identified centrally and consensually by a panel of 3 expert readers (AI vs standard reading)
Time frame: Through study completion, 5 years
Evaluation of the diagnostic performance of the A.I. solution in terms of quality of preparation of the various segments of the digestive tract
Diagnostic performances in quoting small bowel and colonic quality of preparation, as compared toed centrally and consensually quoted by a panel of 3 expert readers
Time frame: Through study completion, 5 years
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.