A comparative study of COVID-19 point-of-care detection across human biofluids using MEMS-based near-infrared spectroscopy and machine learning.

This study explores the use of near-infrared (NIR) spectroscopy in the 1.3–2.6 μm wavelength range, employing a handheld miniaturized microelectromechanical systems (MEMS)-based spectrometer for the rapid, non-invasive detection of COVID-19 from various biofluids. A total of 238 samples—including nasopharyngeal (NSP) swabs, nasal swabs, and saliva—from both COVID-19 positive and negative individuals are analysed. Machine learning algorithms process the spectral data to develop predictive models for the disease classification. Models based on a single biofluid achieve detection accuracies between 75% and 80%, while combining scans from multiple biofluids of the same individual improves accuracy to 88%. The study highlights trade-offs between sample accessibility and diagnostic performance. Overall, the findings demonstrate that NIR spectroscopy serves as a viable low-cost, portable, and rapid point-of-care (POC) solution, with strong potential for scalable mass screening—particularly in resource-limited settings.