COVID-19 and Viral Pneumonia Classification Using Radiomic Features and Deep Learning

The new coronavirus has become the greatest challenge of the 21st century. But since the first cases, much is being discovered about the disease and its effects on the body. Medical imaging, such as X-Rays and CT is widely used to visualize and follow up the patient's clinical picture, especially the effects on the lungs. Although useful, the analysis of this type of image requires some expertise from the radiologist. In less developed countries, the amount of radiologists specialized in chest X-Rays is inadequate, which motivates the development of new technologies to assist clinicians to provide reliable diagnoses. Therefore, this paper addresses the development of a computer-based method to assist in COVID-19 detection among viral pneumonia and health patients through X-Rays images. The proposed method is based on extracting radiomic features and analyzing them using Deep Neural Networks. Experiments following K-Fold Cross-Validation achieved an overall accuracy of 94.98%, a sensibility of 94.89% and an AUC of 99.20%. A benchmark with traditional machine learning algorithms and a binary assessment are also provided. From a multiclass perspective, the analysis and differentiation of COVID-19 and other viral pneumonia reached great results and may assist radiologists in better diagnosing the disease worldwide.