Pneumonia is an acute respiratory infection that causes approximately 2.5 million deaths annually worldwide, with the highest burden occurring in developing countries. Early and accurate diagnosis using chest X-rays is crucial, but requires radiological expertise that is often scarce in resource-limited settings. This study developed an automated pneumonia detection system based on a Deep Convolutional Neural Network (CNN) with an architecture comprising five progressive convolutional blocks. The model was trained using the Chest X-Ray Pneumonia dataset from Kaggle, which contains 5,863 paediatric chest X-ray images. The implementation includes image pre-processing (conversion to greyscale, resizing to 150×150 pixels, and normalisation), real-time data augmentation (rotation ±30°, zoom ±20%, translation ±10%, and horizontal flipping), as well as regularisation techniques including batch normalisation and a dropout layer to reduce overfitting. The network utilises progressively increasing filter sizes (32-64-64-128-256), optimised via RMSprop with an adaptive learning rate scheduling mechanism. Evaluation results on 624 test images show an accuracy of 90.71% with a sensitivity of 91.54% for pneumonia detection. The model achieves a precision of 93% for the pneumonia class and 86% for the normal class, indicating balanced performance. The confusion matrix shows 357 true positives, 209 true negatives, 25 false positives, and 33 false negatives. This study demonstrates that deep learning approaches can serve as effective diagnostic tools for radiologists, particularly in medical centres with limited radiology expertise.
Copyrights © 2026