AI Driven Solution for the Detection of COVID-19 Using X-ray images

Abstract

COVID-19 is a contagious and highly infectious disease which has led to an ongoing pandemic. Researchers and scientists across the world, across various fields, are exploring new methods and approaches to fight against the disease since its outbreak. A study of the COVID-19 infected patients suggests that these patients are affected with the lung infection. In this paper, we have leveraged several deep learning models using the concept of transfer learning. We have also designed a custom convolutional neural network for the purpose of feature extraction and then for effective categorization into pneumonia, covid and normal classes, several classification methods from the machine learning domain such as SVM, Random Forest and softmax regression were utilised. The custom convolutional neural network with the final layer as the dense layer with three units employing softmax activation function achieved a significant accuracy of 94.6 % which was comparable to the accuracy achieved by the transfer learning models. In order to ensure the results are not biased in favour of one class we have utilized a balanced dataset containing 1345 X-ray images for each class - pneumonia, covid, normal in order to demonstrate these experiments.