Proceedings of The World Conference on Innovation in Technology and Engineering Sciences
Detection of Covid-19 in Chest X-ray Image by Using Convolutional Network Trained with Walsh Functions
Muhammed Nur Talha Kılıç, Tamer Ölmez
Covid-19 is a highly contagious disease with devastating problems, comprised of many deaths, heavy costs incurred during the treatment, physiological and psychological problems, and still affects many people all over the world. Several approaches are widely used to either stop or narrow the number of people infected. Chest x-ray images can be also used as a leading indicator once the person gets infected with Covid-19. In image classification, deep neural network structure, including a convolutional neural network in the position of feature extractor and fully connected neural network, is commonly preferred to be able to classify the image among the group. In FCNN, aside from laborious hyperparameter determinations, it also demands high computational load and high memory. The proposed method aims to use a minimum distance classifier with Walsh functions instead of a fully connected neural network. By doing so, many problems such as long training time, high memory requirement coming along with FCNN would be resolved. X-ray images in the dataset have been labeled as Covid-19, lung opacity, normal and viral pneumonia provided by public resources. The proposed small-size model is observed to be able to classify the images at a 92% accuracy rate without benefitting from a highly complex fully connected neural network section.
keywords: Chest X-ray image classification, Covid-19, deep neural networks, Walsh matrix, Convolutional Neural Network.