Detection of COVID-19 from X-Ray Images: A Novel Three-Phase Approach Combining an Ensemble of Customized Convolutional Neural Networks with Feature Extraction and Fusion

Corona Virus Disease 2019 (COVID-19) is a new disease based on the SARS-CoV-2 virus. The virus has caused a worldwide pandemic due to its high infection rate and severity of symptoms. Several methods for detecting the virus exist among which different medical imaging modalities, particularly X-Ray imaging. This imaging modality remains safe and cheaper than others. However, when used to diagnose COVID-19 patients, it showed a higher number of false negatives. We propose a three-phase machine learning approach to detect, from X-Ray images, whether a person is infected with the virus or not. The approach relies on an ensemble of customized convolutional neural networks to extract essential features from input images. The extracted features undergo fusion and are passed on to a classifier for final results. We validate the approach on a set of 3,886 X-Ray images of patients carrying the virus, patients suffering from viral Pneumonia, and healthy persons. When benchmarked against several models published in the literature, our proposed model outperforms them all and achieves accuracy, F1-score, precision, recall and AUC above 99%. Hence, the proposed model have great potential helping medical experts detecting the disease in this type of images.