Addressing Unconventional Hurdles with Machine Learning

[Submitted on 24 Aug 2023]

Download a PDF of the paper titled “The intersection of video capsule endoscopy and artificial intelligence: addressing unique challenges using machine learning” by Shan Guleria and 14 other authors.

Download PDF

Abstract: Introduction: This paper addresses the challenges faced in using artificial intelligence (AI) to improve the practical utility of video capsule endoscopy (VCE). The five challenges identified are: 1) stochastic and artifact-ridden VCE data, 2) cost-intensive interpretation, 3) imbalanced data, 4) computationally cumbersome existing AI models, and 5) clinician hesitancy towards AI models that cannot explain their process. The paper presents various AI-based models that address these challenges, including a convolutional neural network (CNN) for classifying VCE data, a tool for expert annotation, and models based on graph representation and meta-learning. The results show accurate identification of anatomic landmarks and good performance of the different AI models. The goal of producing high performance using lightweight models that improve clinician confidence was achieved.