Stack-Based CNN Approach to Covid-19 Detection

Covid-19, declared as a pandemic by the World Health Organization (WHO), has infected more than 113 million globally across 221 countries. In this work, we propose a method for automatic detection of coronavirus based on analyzing the Chest X-ray images. The dataset used for the study composes of 1200 Covid-19 infected, 1,345 Viral Pneumonia infected and 1,341 healthy patient X-ray images. We use different CNN architectures pretrained on ImageNet dataset, fine tune them to adapt the dataset and use it as feature extractors. We determine the best feature extractor among them, stack them with fully connected layers and employ different classification approaches such as softmax, XGBoost and Support Vector Machines (SVM). The results show that the stacked CNN model with DenseNet169, fully connected layers and XGBoost achieves an accuracy, recall and Fl-score of 99.679% and precision of 99.683%. Hence, the proposed model showcases potential to assist physicians and make the diagnosis process more accurate and efficient.