Chest X-ray Pneumonia Detection using Deep Learning

Pneumonia is a serious lung infection that, as in most cases, requires prompt diagnosis to increase the chances of successful treatment. In this research, deep learning, specifically Convolutional Neural Networks (CNN), is used to perform classification on chest X-ray images by splitting them into two categories - NORMAL (healthy) and PNEUMONIA (infected). To make the dataset more diverse and robust, the data was preprocessed by resizing, normalizing, and augmenting it. A CNN was implemented in which the architecture consisted of convolutional layers with batch normalization, max pooling, global average pooling, and dense layers. The model was trained on the Adam optimizer with binary cross-entropy loss. While the model achieved a training accuracy of 83.32% alongside 74.20% for test accuracy, the model was heavily overfitted when validated and only achieved 62.50% accuracy. Looking at the trends of training and validation loss, along with the confusion matrix, there seemed to be difficulties with correct generalization, most likely as a result of a complex model, an imbalanced dataset, or both. Possible modifications would be to use Transfer Learning with ResNet or EfficientNet, to increase the dataset size, and to conduct the tests in actual clinical environments, which would make the pneumonia detection more reliable and accurate. Keywords —Pneumonia detection, chest X-ray, Image classification