Wireless Capsule Endoscopy for Crohn's Disease Automated Image Classification Using Deep Learning

Despite the diagnosis of Crohn's Disease now transformed through wireless capsule endoscopy (WCE), traditional manual interpretation by gastroenterologists contains substantial restrictions such as high inter-observer inconsistency, time dedication, and potential oversight of important clinical signs. The paper proposes an automated image classification system founded on deep learning methods specifically, Convolutional Neural Network (CNN) for feature extraction and Long Short-Term Memory networks (LSTM) for temporal sequence examination. The technique applies Transfer Learning to improve model performance with minimal labeled data, thus facilitating precise identification of CD-related abnormalities. Strong testing on an endoscopist-annotated dataset indicates the proposed system substantially surpasses existing systems about accuracy of 94.5%, sensitivity of 92.0%, and specificity of 96.0%. Promising enhanced patient outcomes in the therapy of Crohn's Disease, the automated method significantly decreases interpretation time, minimizes inter-observer inconsistency, and increases diagnostic accuracy.