Automatic Identification of COVID-19 in Chest X-Ray Images Based on Deep Features and Machine Learning Models

ABSTRACT

In 2020, the novel coronavirus (COVID-19), spread around the world and became a pandemic. It is diagnosed by a Real-Time Reverse Transcriptase Polymerase Chain reaction (RT-PCR) test, which requires a specialized laboratory to confirm the presence of the virus. Due to the insufficient availability of these labs, medical images have been used as an alternative diagnosis, being the most easily available and least expensive option the Chest X-Ray. As COVID-19 infected patients display very similar respiratory affections like other kinds of pneumonia, distinguish them is difficult even for experienced radiologists. In this paper, two popular deep learning architectures are used to extract deep features, which are then used for training multi-class classification machine learning models to distinguish COVID-19 from healthy, bacterial, and other viral pneumonia infections. The evaluation was performed on a dataset of 7732 images, including 1575 healthy patients, 2801 diagnosed with bacterial pneumonia, 1493 with a viral (no COVID) infection, and 1863 subjects with COVID-19 confirmed diagnosis. The general area under the ROC curve was between 93 % ± 2 % for general categories;and 99 % ± 1 % with a sensitivity of 83 % ± 2 % to identify COVID-19 infected patients. © 2022, Springer Nature Switzerland AG.