Pulmonary Disease Detection And Classification Using Feature Wise Attention Residual Expansion Network Using Chest X-Ray Images

Abstract

Chest X-Ray (CXR) images are one of the most widely utilized medical imaging techniques to diagnose pulmonary diseases due to its effectiveness and accessibility. However, accurately classifying pulmonary disease is challenging due to subtle differences and overlapping features that leads to suboptimal performance. In this research, Feature wise Attention Residual Expansion Network (FARENet) is proposed in Multi Layer Perceptron (MLP) to detect and classify pulmonary disease accurately. The expansion and projection layers in MLP enhance feature representation through increasing dimensionality for capturing complex patterns and then minimizing to retain only most appropriate information that enhance efficiency. Feature wise Channel Attention Mechanism (CAM) refines this process via assigning importance to significant features that ensure better discrimination between pulmonary disease classes. Moreover, residual skip connection provides effective gradient flow and conserve essential information from CAM that avoids model degradation. These components enhance classification accuracy, learning ability, and minimize redundancy. Hence, proposed method achieves high accuracy of 99.15%, 95.23%, and 99.41% on CXR, PulmoNet, and COVID-19 datasets compared to existing methods like InceptionResNetV2.