COVID-19 classification using thermal images: Thermal images capability for identifying COVID-19 using traditional machine learning classifiers

Abstract

Medical images have been proposed as a diagnostic tool for SARS-COV-2. The image modality more investigated on this subject is computed tomography (CT), however it has some disadvantages: it uses ionizing radiation, requires unique installations along with a complicated process limiting the number of possible tests per equipment, and the economic costs can be prohibitively high for screening a large population. For these reasons, the aim of this study is to investigate thermal images as an alternative modality for diagnosis of COVID-19. The methodology used in this study consisted of using radiomics and moment features extracted from six images obtained from thermal video clips in which optical flow and super resolution were used, these features were classified using traditional machine learning methods. Accuracies were in the range of 0.433-0.524. These first results conducted on thermal images suggest that the use of this type of image modality is unlikely to be favorable for COVID-19 detection.