An Attention-Enhanced Hybrid Deep Learning Model Based on VGG16 and VGG19 for Pneumonia Detection from Chest X-ray Images

Abstract

Pneumonia is one of the most dangerous respiratory diseases and could be life-threatening if not promptly diagnosed and treated. In addition, pneumonia is an infectious disease, and missing a case poses a significant risk to the community. Conventionally, doctors rely on chest X-ray (CXR) images to examine the lungs and detect abnormalities associated with pneumonia. The development of artificial intelligence (AI), especially deep learning (DL) algorithms, can assist doctors in diagnosing the disease more quickly and accurately. This study proposes a hybrid DL model that combines two convolutional neural networks (CNNs), VGG16 and VGG19, with an attention mechanism to enhance pneumonia detection from CXR images. By integrating the lightweight structure of VGG16 with the deeper feature extraction of VGG19 and directing focus to key pathological regions through attention, the model achieves improved diagnostic performance. Evaluated on a public pediatric CXR dataset, the proposed model outperforms VGG16, VGG19, DenseNet121, and InceptionV3 in all major metrics: 89.10% accuracy, 86.42% precision, 91.82% F1-score, and 97.95% recall. The high recall rate is particularly significant in minimizing false negatives, which is critical in clinical contexts to prevent missed pneumonia cases. Despite having the highest parameter count among the compared models, it maintains a fast inference time of 33.48 ms per image, supporting real-time clinical application.