Highly accurate multiclass classification of respiratory system diseases from chest radiography images using deep transfer learning technique

Chest X-ray radiographic (CXR) imaging aids in the early and accurate diagnosis of lung disease. The diagnosis process can be automated and accelerated by analyzing chest CXR images with artificial intelligence tools, particularly Convolutional Neural Network (CNN). Due to few medical images have been labeled, the most significant obstacle is utilizing these images accurately for diagnosing and tracking disease progression, and accordingly, the difficulty of automating the classification of these images into positive and negative cases. To address this issue, a deep CNN model was proposed to classify respiratory system diseases from X-ray images using a transfer learning technique based on the EfficientNetV2 model that acts as a backbone to enhance the efficacy and accuracy of Computer-Assisted Diagnosis (CAD) performance. Moreover, the latest data augmen-tation methods and fine-tuning for the last block in the convolutional base have also been carried out. In addition, Grad-CAM is used to highlight the important features and make the deep learning model more comprehensible. The proposed model is trained to work on the triple classification, COVID-19, normal, and pneumonia. It uses CXR images from three publicly accessible datasets. The following performance was achieved on the testing set: sensitivity = 98.66 %, specificity = 99.51 %, and accuracy = 99.4 %. Thereby, the proposal outperforms the four most recent classification techniques in the literature.