CLASSIFICATION OF CHEST X-RAY IMAGES USING CAPSULE NETWORK

Abstract

The classification of chest X-ray (CXR) images plays a critical role in the early detection and diagnosis of various pulmonary diseases. Conventional deep learning approaches, particularly Convolutional Neural Networks (CNNs), often face limitations in preserving spatial hierarchies and handling variations in orientation, scale, and viewpoint present in medical imaging. To address these challenges, this study proposes the use of Capsule Networks (CapsNets) for effective chest X-ray image classification. CapsNets utilize dynamic routing and vector-based feature representation to capture part–whole relationships more efficiently than traditional CNNs, enabling improved robustness and interpretability. A curated dataset of labeled CXR images was used to train and evaluate the model, and experimental results demonstrate that the CapsNet-based approach achieves superior accuracy and generalization for detecting abnormalities compared to baseline CNN architectures. The findings highlight the potential of Capsule Networks as a powerful and reliable technique for medical image analysis, enhancing diagnostic performance and providing more meaningful feature representation for clinical decision support.