Deep Learning Methods for Pneumonia Detection Using ConvNeXt Architecture

Abstract

Pneumonia remains a major respiratory infection with high mortality rates, especially in regions with limited access to medical specialists. Early and accurate diagnosis plays a critical role in reducing fatal outcomes and improving patient management. Chest X ray imaging is widely used as a primary diagnostic modality, yet interpretation relies heavily on experienced radiologists, whose availability is often insufficient to meet clinical demand. This condition motivates the development of automated pneumonia detection systems based on artificial intelligence. This study investigates the application of deep learning for pneumonia classification using chest X ray images, with a focus on the ConvNeXt architecture. ConvNeXt represents a modern neural network design that integrates structural advantages from Vision Transformers with the efficiency of traditional Convolutional Neural Networks, enabling strong feature extraction while maintaining computational efficiency. The research evaluates multiple ConvNeXt variants, including Tiny, Small, Base, and Large, to analyze the relationship between model complexity and classification performance. ResNet50 is employed as a baseline model to provide a fair comparative assessment against a widely used convolutional architecture. Model evaluation uses accuracy, precision, recall, and F measure to ensure balanced measurement across different classification outcomes and class distributions. Experimental results indicate that ConvNeXt Tiny achieves the highest overall performance, reaching an accuracy of 97.69 percent while using a relatively low number of parameters. This outcome highlights the efficiency of lightweight architectures for medical image analysis tasks. The findings demonstrate that ConvNeXt Tiny delivers strong discriminative capability with reduced computational requirements, making it suitable for deployment in resource constrained clinical environments. This study contributes evidence supporting the effectiveness of modern deep learning architectures for automated pneumonia detection and provides insight into model selection for practical medical imaging applications.