Attention Mechanism for Pulmonary Disease Detection in X-Ray Images

This paper proposes an attention mechanism to classify lung X-ray radiographs as normal, lung opacity, as well as Viral Pneumonia. The proposed architecture integrates convolutional layers, residual layers, inception layers, and a multilayer attention mechanism. We have used augmentations of Gaussian blur and the addition of Salt and Pepper noise to increase variations in the dataset. This proposed network improves the classification accuracy of the Lung Disease Dataset, improving traditional models like VGG16, VGG19, DenseNet, and InceptionV3. This architecture achieves superior accuracy by effectively capturing spatial hierarchies and focusing on the most relevant features.