Vascular Lesions

Vascular lesions are frequently found during capsule endoscopy exams performed for obscure gastrointestinal bleeding. However, determining the culprit lesion in these cases is often difficult. Our group developed a deep learning algorithm for automatic detection of vascular lesions (red spots, angiectasia, varices) and their categorization according to a validated scale to predict their bleeding potential (Saurin’s scale). Our convolutional neural network detected these lesions and differentiated their bleeding potential with an overall accuracy of 94%, with a sensitivity of 92% and a specificity of 96%. Moreover, our neural network achieved a high image processing speed of 145 frames/second. This work has been accepted for publication as an original article in Annals of Gastroenterology.
This the first artificial intelligence algorithm providing both detection of vascular lesions and categorization of their bleeding potential according to a validated scale. This is important as it may significantly impact further diagnostic and therapeutic strategies. The link for the complete article will posted here soon.

Accepted manuscript // March 2021