Deep Residual Learning for Multi-Class Diagnostics in Capsule Endoscopy

Gastrointestinal (GI) diseases encompass a wide range of conditions that impact various parts of the digestive system, with symptoms ranging from mild discomfort to lifethreatening complications. Wireless capsule endoscopy (WCE) has emerged as a non-invasive and patient-friendly method to visualize the GI tract, especially for detecting abnormalities such as ulcers and arteriovenous malformations (AVMs). In this study, we propose a deep learning-based classification approach for analyzing WCE images and identifying three GI conditions: Normal, Ulcer, and a new class, AVM. Its classification is very important as it improves the diagnosis of gastrointestinal diseases. The King Abdulaziz University Hospital-Capsule (KAUHC) dataset, captured using the OMOMWCE system, was used for training and evaluating five ResNet variants. Images were preprocessed and augmented to enhance generalization, and models were trained using transfer learning and optimized with a standardized configuration. Evaluation metrics, including accuracy, precision, recall, and F1-score, demonstrated that the ResNet-34, ResNet-50, and ResNet-101 models achieved the best overall performance, with classification accuracy reaching up to $\mathbf{9 9. 4 \%}$. A comparative analysis with traditional machine learning models (e.g., Decision Trees, KNN, SVM) revealed that deep learning significantly outperformed these methods in both consistency and classification accuracy. These findings highlight the effectiveness of integrating capsule endoscopy with deep learning to enhance diagnostic precision in gastroenterology and support more accurate, non-invasive medical decision-making.