Explainable CNN for Automated Lung Opacity Detection Using Fine-Tuned VGG-19 with Grad-CAM

<title>Abstract</title> <p> Pneumonia is a serious respiratory disease and one of the leading causes of death among children under five years of age. Manual interpretation of chest X-ray images by radiologists is time-consuming and may lead to diagnostic errors. This study proposes an automated lung opacity detection system based on transfer learning using a fine-tuned VGG19 architecture. The convolutional backbone of the pretrained VGG19 model is frozen, while a new binary classification layer is trained to distinguish between normal and opacity chest X-ray images. A publicly available Kaggle dataset containing two classes, Normal and Opacity, is used for training and evaluation. The proposed model achieves an overall accuracy of <bold>93%</bold> and demonstrates high recall for opacity cases, making it suitable for supporting clinical diagnosis. To improve transparency and trustworthiness, Grad-CAM is integrated to visualize the regions responsible for the model's predictions through heatmaps. The experimental results indicate that the model shows stable learning behavior without significant overfitting and can serve as a reliable computer-aided diagnostic tool for pneumonia detection. </p>