Detecting Covid-19 Respiratory Markers with Ordinary Mobile Phones

Abstract

Background: Covid-19 is a global tragedy where robust disease detection technologies remain inaccessible to billions of underserved people, many of whom live in remote areas. The purpose of this clinical study was to evaluate the feasibility of detecting Covid-19 with ordinary (unmodified) mobile phones by using Tele-stethoscopeA, a technology developed by Fleming Scientific, to characterize respiratory airflow. Method: The study took place in a hospital environment where patients were tested, on admission, with CoviPath™-19 RT-PCR Kit or equivalent and verified to be positive or negative for Covid-19. Each patient’s respiration was then recorded just above the collarbone (supraclavicular fossa) using an auscultation technique called egophony. Two android phone brands were used to record the respiratory acoustics with a free open source voice recorder. From 260 total patient recordings, 210 were randomly selected for training a diagnostic model and 50 were held out for model testing. Results: The diagnostic model achieved 84% sensitivity and 85% specificity among the training set. Among the test set, the model achieved 80% sensitivity and 72% specificity. Conclusion: Although a number of study factors worked against achieving better test set performance, these results support the feasibility of detecting Covid-19 with ordinary phones. In broad distribution, the technology could be an important aid to personal and public healthcare management of Covid-19. It could extend robust testing to underserved patients and situations globally. The same technology could, possibly, be applied to detect remotely other diseases that present in bronchial acoustics.