A Deep Learning Model for Detection of Lung Pathologies by Image Processing of X-ray Images

Lung pathologies including viral pneumonia, covid-19, lung opacities and tuberculosis are among the most widely spread lung infections. The primary detection of these lug pathologies is done by the examination of the chest radiographs from which the structural abnormalities are observed to diagnose the lung disease. With the increase in cases of lung pathologies and the reduced number of radiologists and diagnostic labs available for x-ray examination, diagnosing these diseases has become worrisome. In this paper, an automated deep learning model designed using Matlab 2021a which proved very helpful for accurate and quick radiograph examination has been proposed. The data obtained from the online database Kaggle is trained using transfer learning approach for training a pre-trained convolution neural network ResNet50 model that can efficiently classify x-ray images into 4 distinguished classes of viral pneumonia, covid-19, lung opacities and normal radiographs based on structural changes in the lungs caused by the respective pathology. 80% of data is used for training the model and 20% for model testing. The trained model classified the radiographs with a high accuracy of 94% which provide us with a highly accurate model for radiograph examination. The results were displayed in the form of a confusion matrix, occlusion sensitivity figure and class activation map. ResNet50 model showed better performance as compared to excessively opted deep transfer learning models. Graphical user interface is developed using this approach to make the model user-friendly.