Attention-Enhanced Transfer Learning for Emphysema Classification Using CBAM-Augmented ResNet-50

Respiratory diseases, particularly emphysema, rank among the most severe public health problems globally. Therefore, there is a need for accurate and effective diagnostic methods. This work proposes a new deep architecture that combines transfer learning using Convolutional Block Attention Modules to enhance ResNet-50 performance in binary classification on chest X-ray images. The dataset was from the NIH Chest Xray, which underwent preprocessing for class balancing before training and testing. Preprocessing included denoising, enhancing the image, and data augmentation to improve the quality and diversity of the data. The proposed architecture uses CBAM to dynamically refine feature maps along spatial and channel dimensions to focus the model on clinically relevant regions. Evaluations using metrics such as accuracy, precision, recall, and F1-score show the model’s efficiency by achieving an accuracy of 92.81% and a balanced F1-score of 90%. This study has shown the potential of attention-enhanced deep-learning frameworks to advance medical diagnostics.