Long-term device stability for Raman spectroscopy

Abstract

Long-term stability of Raman setups is one of the critical criteria for using Raman spectroscopy in real-world applications. Substantial differences from long-term drifts of a device can largely reduce the reliability of the technology and lead to serious consequences in scenarios such as disease diagnostics. A systematic investigation of long-term device stability is urgently needed to understand the device-related variations and to help improve the situation. In this study, 13 substances were measured as quality control references weekly for 10 months on a Raman device as quality control references to investigate instrumental stability over time. The 13 substances were selected to be stable and to cover a wide range of standards, solvents, lipids, and carbohydrates. Approximately 50 Raman spectra of each substance were acquired per measurement day. A data pipeline was constructed to discover the variability (i.e., instability) of the device for the covered time window. Therein, the stability of the measurement was benchmarked from multiple perspectives, including the intensity variations, the correlation coefficients, the clustering, and the classification. The results suggested the device variability to be more random than systematic. Nonetheless, we demonstrated the possibility of decreasing the variations from the data via computational methods. In particular, we estimated the spectral variations by a network adapted from the variational autoencoder (VAE) and suppressed them from the measured data by the extensive multiplicative scattering correction (EMSC) method. This could improve the prediction of independent measurement days for three representative classification tasks.