Utilization of a Deep Convolutional Neural Network for the Binary Classification of Chest X-Ray Pneumonia

Abstract

Pneumonia remains a significant global health concern, necessitating efficient diagnostic tools. This study presents a novel Convolutional Neural Network (CNN) architecture, CuDenseNet, designed for the binary classification of Chest X-Ray (CXR) images as either having pneumonia or normal (healthy). Unlike models that rely on transfer learning from pre-trained architectures, CuDenseNet is trained from scratch and incorporates three parallel DenseNet paths of varying depths, enhancing feature extraction and classification accuracy. The model was evaluated on a combined dataset of 11,708 CXR images, achieving exceptional performance metrics of 99.1% accuracy, 99.7% precision, 99.1% recall, and an AUC of 99.7%. The comparative analysis demonstrates that CuDenseNet outperforms state-of-the-art pre-trained models such as VGG19 and ResNet50 while providing superior adaptability. These results underscore the potential of CuDenseNet as a robust and reliable tool for automated pneumonia diagnosis, with significant implications for clinical applications and future research in medical imaging.