Intelligent COVID-19 screening platform based on breath analysis

The spread of coronavirus disease 2019 (COVID-19) results in an increasing incidence and mortality. The typical diagnosis technique for severe acute respiratory syndrome coronavirus 2 infection is reverse transcription polymerase chain reaction, which is relatively expensive, time-consuming, professional, and suffered from false-negative results. A reliable, non-invasive diagnosis method is in urgent need for the rapid screening of COVID-19 patients and controlling the epidemic. Here we constructed an intelligent system based on the volatile organic compound (VOC) biomarkers in human breath combined with machine learning models. The VOC profiles of 122 breath samples (65 of COVID-19 infections and 57 of controls) were identified with a portable gas chromatograph-mass spectrometer. Among them, eight VOCs exhibited significant differences (p< 0.001) between the COVID-19 and the control groups. The cross-validation algorithm optimized support vector machine (SVM) model was employed for the prediction of COVID-19 infection. The proposed SVM model performed a powerful capability in discriminating COVID-19 patients from healthy controls, with an accuracy of 97.3%, a sensitivity of 100%, a specificity of 94.1%, and a precision of 95.2%, and anF1 score of 97.6%. The SVM model was also compared with other common machine models, including artificial neural network,k-nearest neighbor, and logistic regression, and demonstrated obvious superiority in the prediction of COVID-19 infection. Furthermore, user-friendly software was developed based on the optimized SVM model. The developed intelligent platform based on breath analysis provides a new strategy for the point-of-care screening of COVID and shows great potential in clinical application.