Enhanced Pneumonia Detection from Chest X-rays via VGG-16 and Self-Attention Mechanisms

Abstract

Early and accurate detection of pneumonia in chest X-rays is critical for effective treatment, especially in resource-constrained healthcare settings. Manual diagnosis is time-consuming and prone to variations, underscoring the need for robust automated approaches. This study addresses the challenge of improving the diagnostic accuracy and interpretability of deep learning models for pneumonia detection from chest radiographs. The method proposes a novel deep-learning framework that combines transfer learning using a pre-trained VGG16 model with a self-attention-enhanced convolutional architecture. The VGG16 backbone extracts low-level visual features, while the self-attention mechanism highlights clinically relevant lung regions, improving spatial focus during classification. The proposed model leverages the VGG16 backbone to extract low-level visual features, while a self-attention mechanism enhances spatial focus by emphasizing clinically significant lung regions. The VGG16 model, guided by attention, achieved 97% accuracy, precision, and recall in pneumonia detection. Also, Grad-CAM visualizations improved interpretability and model performance compared to baseline CNNs and pre-trained architectures. The integration of a self-attention mechanism into a transfer learning framework significantly improves both the performance and interpretability of pneumonia detection models in chest X-rays. This approach closely replicates the spatial reasoning of human experts and offers a scalable solution for clinical deployment. The results indicate that attention-enhanced deep learning architectures are well-suited for medical imaging tasks, particularly in resource-constrained settings where diagnostic expertise may be limited.