Early Detection of Pneumonia using Image Classification with Convolutional Neural Networks

Pneumonia is an important global source of morbidity and mortality, and it is vital in the timely diagnosis and treatment of pneumonia. Traditional diagnostic methods, like a chest X-ray, rely heavily on radiologists to analyze radiographs. Thus, diagnosis can take time and is prone to human error. In this paper, we introduce a system that automates the early detection of pneumonia from chest X-ray images by using a Convolutional Neural Network (CNN). This work leverages the publicly available Chest X-ray dataset that contains 5,863 labeled images of normal and pneumonia cases. The images were pre-processed by normalizing the images, resizing the images to 224×224 pixels, and applying data augmentation techniques (e.g., rotation and flipping). We compared different deep learning models, including a custom various CNN architecture and pre-trained networks (e.g., VGG16 and ResNet50) that used transfer learning. Out of the models evaluated, VGG16 demonstrated the best performance with test accuracy of 93.6%, precision of 94.1%, recall of 92.8%, and F1-score of 93.4%. The ROC curve analysis produced an AUC (area under the curve) of 0.96, which indicates that our model was robust. These preliminary results show that machine learning specifically CNNs, can be used to support healthcare professionals in the early identification of pneumonia and therefore lead to improved patient outcomes and less delay to diagnosis.