Automated Pneumonia Detection in Healthcare Using Efficient Deep Learning

Abstract

Accurate and timely pneumonia detection by using chest radiographs plays an important role in reducing death rates and also supports clinical decision-making. In recent years, medical image analysis has advanced considerably and deep learning techniques have made a significant contribution. However, finding a model that provides both high accuracy and low computational cost is still difficult. MobileNetV2 is adopted as the core architecture for binary classification of normal and pneumonia cases. This proposed approach is compared with the traditional Convolutional neural network and the well-known VGG16 model. To keep the evaluation consistent all models are trained and validated under the same settings. Metrics used for model performance evaluation are Accuracy, F1-score, confusion matrix and AUROC. Experimental results of MobileNetV2 are accuracy of 93.75%, F1-score of 93.33% and AUROC of 1.00, indicating that the proposed model outperforms VGG16 having 81.25% accuracy, 84.21% F1-score and an AUROC of 98.44%, and baseline CNN having 75.00% accuracy, 80% F1-score and an AUROC of 98.44%. MobileNetV2 offers an adequate compromise between the performance of classification and computational efficiency. The results confirm that lightweight deep learning architectures can deliver high diagnostic performance for pneumonia detection and may serve as practical solutions for real-world and resource-constrained clinical deployment.