Transfer Learning-Driven Binary Classification of Chest X-ray for Pneumonia Using Deep Convolutional Architectures

Abstract

Pneumonia is the most common infectious cause of lung inflammation, resulting from the presence of viruses or bacteria in the microscopic air sacs. In recent years, artificial intelligence, particularly deep learning, has gained significant traction in the field of medical imaging. Early detection and diagnosis of pneumonia using chest X-rays can support radiologists in clinical decision-making and subsequent treatment planning, thereby helping to reduce mortality rates. In this study, a chest X-ray dataset was used to classify images for pneumonia detection using transfer learning with four deep learning architectures: Visual Geometry Group 16 (VGG16), DenseNet201, Inception V3, and Inception ResNet V2. Each model was developed, implemented, and evaluated, and the results were analyzed and compared. The best-performing model was identified based on accuracy and minimal loss, demonstrating promising proficiency and robustness. The dataset employed in this study was obtained from Kaggle. The methodology involved data collection, preparation, preprocessing, and visualization. Models were developed and implemented using the Jupyter Notebook editor via Google Colaboratory. Among the architectures evaluated, DenseNet201 achieved superior performance compared to VGG16, Inception V3, and Inception ResNet V2 in terms of classification accuracy.