Deep Learning-Based Binary Classification of Pneumonia from Chest X-Ray Images Using PneumoniaMNIST: Development and Evaluation of a Convolutional Neural Network

Pneumonia remains an important respiratory condition for which timely identification can support appropriate clinical assessment and treatment. This study presents the development and evaluation of a convolutional neural network (CNN) for binary classification of pneumonia from chest X-ray images using the PneumoniaMNIST dataset. The dataset comprised 5,856 labelled grayscale chest X-ray images divided into predefined training, validation, and test sets containing 4,708, 524, and 624 images, respectively. Images were normalised and processed at a resolution of 28 × 28 pixels, while conservative data augmentation was applied during training. The proposed CNN incorporated three convolutional blocks with batch normalisation and max pooling, followed by a fully connected layer, dropout regularisation, and a sigmoid output layer for binary classification. Model training employed the Adam optimiser, binary cross-entropy loss, learning-rate adaptation, and early stopping. Evaluation on the independent test set demonstrated an accuracy of 90.87%, precision of 87.93%, sensitivity (recall) of 98.97%, specificity of 77.35%, F1-score of 93.12%, and receiver operating characteristic area under the curve (ROC-AUC) of 97.30%. The high sensitivity indicates that the model successfully identified the majority of pneumonia-positive cases within the test dataset, while the comparatively lower specificity indicates a greater tendency to misclassify some normal cases as pneumonia. These findings demonstrate the potential of convolutional neural networks for automated chest X-ray classification while highlighting the importance of balancing sensitivity and specificity in medical image analysis. The proposed model is intended as a research-based decision-support approach and has not been clinically validated as a standalone diagnostic system.