P445 Wireless capsule endoscopy reading support for Crohn’s disease patients using deep neural networks with an active learning method
Résumé
Artificial intelligence (AI) systems, in particular neural networks (NN), have been developed to detect lesions in wireless capsule endoscopy (WCE) in patients with Crohn’s disease (CD) in order to ease their reading. Despite acceptable performance on individual images, these tools do not analyze entire videos and require a large amount of labeled data. Active learning (AL) methods have already been used, proving their ability to limit labeling cost without reducing NN performance. The aim of this study was to evaluate several AL models in order to train an AI system able to differentiate normal and pathological entire videos with a limited labeled dataset.