Rapid reagent free COVID19 detection using MEMS based FTIR spectroscopy and machine learning in NIR and MIR regions

Abstract

Abstract This study presents rapid, reagent-free detection of COVID-19 using miniaturized MEMS-based Fourier-transform infrared (FTIR) spectrometers integrated with machine learning models. Two portable spectrometers analyze 363 nasopharyngeal swab samples stored in viral transport medium (VTM). The first spectrometer covers the near-infrared (NIR) region (1.3–2.6 μm) and second spectrometer extends from the near infrared to the mid infrared (MIR) region (1.75–4.0 μm). The NIR system uses a transmission configuration, while the one extended to the MIR performs attenuated total reflectance (ATR) measurements on both wet and dried samples. Spectral data undergo preprocessing and analysis using interval partial least squares discriminant analysis (iPLS-DA), with model training and evaluation conducted through Monte Carlo cross-validation. The MIR wet sample model achieves a diagnostic performance with 79% accuracy, a 98% sensitivity, and an area under the curve (AUC) of 0.8. The MIR dry sample model achieves an 80% accuracy and an AUC of 0.79, while the NIR model reaches 66% accuracy and an AUC of 0.64. Spectral features appear in the Amide A and B regions in the MIR range, and in the C–H overtone bands in the NIR range. The full measurement process, including sample handling, completes in under six minutes, supporting its suitability for real-time, point-of-care (POC) testing.