Utilized CNN Model for Lung Diseases Detection

According to the World Health Organization (WHO), the leading causes of mortality in the globe are pneumonia, COVID-19, tuberculosis, and pneumothorax. Therefore, early diagnosis and successful treatment of lung illnesses depend critically on the early discovery of these conditions. Moreover, common symptoms include coughing, sneezing, fever, and shortness of breath. Several diagnostics, including Monteux tuberculin skin test (TST), antigen tests, molecular testing (RT-PCR), and complete blood count (CBC), are required to detect them. However, they are time-consuming methods with a 20% error rate and an 80% sensitivity. Therefore, with the assistance of a physician, radiographic tests such as computed tomography (CT) and an X-ray are performed to discover lung illnesses. However, with chest X-ray or CT-scan images, the danger of various lung illnesses’ diagnosis overlapping characteristics is a concern. To properly categorize the lung diseases, such a method must be automated. In object classification and localization tasks, state-of-the-art classifiers built on convolutional neural networks (CNNs) achieve great accuracy. The CNN and the rat swarm optimizer (RSO) serve as the foundation for the suggested hybrid approach. To achieve a significant level of diagnostic accuracy, RSO is utilized for identifying the ideal values for the EfficientNet-B4 model architecture’s hyperparameters. This paper proposes the optimized EfficientNet-B4 architecture for multi-class diagnosis of tuberculosis (TB), pneumonia, lung opacity (LO), and, most recently, COVID-19. The results demonstrate that the proposed model achieves superior performance in lung disorder detection. When predicting lung disorders from Chest X-ray (CXR) images, the suggested model achieves 97.84% accuracy, 98.23% sensitivity, 99.67% specificity, 98.3% precision, and 98.27% F1 Score. Furthermore, the suggested hybrid approach performance is compared to that of various CNN architectures, including Inception v3, Densenet-201, VGG16, and VGG19. Finally, using an optimized EfficientNet-B4 model, we provide a robust and reliable method for identifying lung diseases. This method has the potential to help healthcare specialists discover and diagnose lung problems earlier, resulting in better patient outcomes and better healthcare practices.