Abstract:
Deep learning (DL)-based diagnosis of respiratory illnesses has the ability to detect diseases early. However, the success of DL algorithms is dependent on the datasets, therefore developing an algorithm that works with both CT and X-ray pictures is critical. As a result, this study focuses on the creation of a DL framework to improve the classification of infectious disorders such as coronavirus and pneumonia.This projectproposes a new DL framework for analyzing lung diseases including COVID-19 and pneumonia from chest CT scans and X-ray (CXR) images. This framework is termed optimized DenseNet201 for lung diseases (LDDNet).The proposed LDDNet is developed using additional layers of pooling, dense,and dropout layers, as well as batch normalization to the base DenseNet201 model. Next, two datasets of lung diseases are formed from separate open-access sources. One is a CT scan dataset containing 1043 images and the other is an X-ray dataset comprising 5935 images of COVID-19 and pneumonia-affected lungs and healthy lungs. The performance of the proposed LDDNet is evaluated with the performance of ResNet152V2 and XceptionNet model, where for the sake of real evaluation the proposed framework is implemented for these two models also. The performance of the models is analyzed for Adam, Nadam, and SGD optimizers. Among all these three models, the LDDNet provides the best results for both CT scan and CXR datasets for Nadam optimizer. For CT scan images, the LDDNet shows aCOVID-19 positive classification accuracy of 99.36%, the precision of 100%, recall of 98%, and 99% f1-score. For X-ray images, LDDNet provides 99.55% accuracy, 73% recall, 100% precision, and 85% f1-score while implementing Nadam optimizer in detecting COVID-19-affected patients. For a given set of parameters, the performance of LDDNet is found to be better than the existing algorithms of ResNet152V2 and XceptionNet.The proposed DL methods identify lung disorders relatively effectively and thus can contribute to the prevention of lung disease progression.