Machine-Aided Detection of SARS-CoV-2 from Complete Blood Count

Abstract

The current gold standard for SARS-CoV-2 detection methods lacks the functionality to perform population screening. Complete blood count (CBC) tests are a cost-effective way to reach a wide range of people – e.g. according to the data of the Central Statistical Office of Poland from 2016, there are 3,000 blood diagnostic laboratories in Poland, and 46% of Polish people have at least one CBC test per year. In our work, we show the possibility of machine detection of SARS-CoV-2 virus on the basis of routine blood tests. The role of the model is to facilitate the screening of SARS-CoV-2 in asymptomatic patients or in the incubation phase. Early research suggests that asymptomatic patients with COVID-19 may develop complications of COVID-19 (e.g., a type of lung injury). The solution we propose has an F1 score of 87.37%. We show the difference in the results obtained on Polish and Italian data sets, challenges in cross-country knowledge transfer and the selection of machine learning algorithms. We also show that CBC-based models can be a convenient, cost-effective and accurate method for the detection of SARS-CoV-2, however, such a model requires validation on an external cohort before being put into clinical practice.