Scat-NET: COVID-19 diagnosis with a CNN model using scattergram images

Tuncer, Seda Arslan; Ayyıldız, Hakan; Kalaycı, Mehmet; Tuncer, Taner

The acute respiratory syndrome COVID-19 disease, which is caused by SARS-CoV-2, has infected many people over a short time and caused the death of more than 2 million people. The gold standard in detecting COVID-19 is to apply the reverse transcription polymerase chain reaction (RT-PCR) test. This test has low sensitivity and produces false results of approximately 15%–20%. Computer tomography (CT) images were checked as a result of suspicious RT-PCR tests. If the virus is not infected in the lung, the virus is not observed on CT lung images. To overcome this problem, we propose a 25-depth convolutional neural network (CNN) model that uses scattergram images, which we call Scat-NET. Scattergram images are frequently used to reveal the numbers of neutrophils, eosinophils, basophils, lymphocytes and monocytes, which are measurements used in evaluating disease symptoms, and the relationships between them. To the best of our knowledge, using the CNN together with scattergram images in the detection of COVID-19 is the first study on this subject. Scattergram images obtained from 335 patients in total were classified using the Scat-NET architecture. The overall accuracy was 92.4%. The most striking finding in the results obtained was that COVID-19 patients with negative RT-PCR tests but positive CT test results were positive. As a result, we emphasize that the Scat-NET model will be an alternative to CT scans and could be applied as a secondary test for patients with negative RT-PCR tests.