Comparison of signal calibration methods in the problem of Raman spectroscopy.

Abstract

In the modern world, spectroscopy methods are widely used for non-contact rapid analysis of the composition of liquids [1]. The presence of peaks of natural frequencies of vibrations allows one to accurately determine the presence of certain substances [2], and the general sensitivity of spectral shapes, in particular, sensitivity of the shape of the valence band of water, allows us to draw additional conclusions about the presence and concentrations of salts that do not have their own peaks [3].
Modern machine learning (ML) methods of data analysis, such as neural networks, are widely used to analyze spectral data [4], including multicomponent solutions, obtaining results, which were not able to be obtained with calibration curves. Unfortunately, the downside of these capabilities is the high requirements for data quality - the absence of baseline drift effects, parasitic illumination, and changes in the position of natural vibration peaks. At the same time, spectral methods, having high sensitivity, have low stability of operation: change in the measurement conditions, change in the alignment of the setup, the presence of impurities in solutions lead to changes in the spectral curve, which can significantly affect the results and the operation of ML algorithms. Due to that, in practice, spectral methods without using ML are used only to obtain integral characteristics [5] (for example, the presence of a certain pollutant or an estimate of the total salinity).
In this study, we compare various popular calibration techniques recommended to eliminate sample variability: general correction of the baseline and of the slope of the spectrum [6]; transition to the principal components space [7] and elimination of the main components from data; using partial least squares components to determine components not related with concentration changes [8]; mapping original data to shape space to eliminate translation and scale factors [9]; domain regularized component analysis to obtain new samples relative to reference samples [10]. Then a conventional “shallow” ANN model was implemented to correct the data and to solve the inverse problem.
The original reference dataset consisted of 4445 samples with dissolved 10 inorganic salts with various concentrations: MgSO4, Mg(NO3)2, LiCl, LiNO3, NH4F, (NH4)2SO4, KHCO3, KF, NaHCO3, NaCl. More information about the dataset is given elsewhere [11]. Another “new” datasets that should be corrected and have potential problems (differences from the initial one) were also recorded during real observations. There are several scenarios of alterations in the measurement procedure to obtain these altered datasets: readjustment of the laser equipment setup, change of the cuvette in which the test substance is located, preparation of a solution based on another commercial supply of similar salts substances.
The result of the study is a clear illustration of the appearance of changes in the spectra under different measurement conditions and a discussion of the most effective methods to get stable results.
This study has been performed with financial support from the Russian Foundation for Basic Research, project no. 19-01-00738.
References:
1. Carpenter, N. G. and Pletcher, D., “Amperometric method for the determination of nitrate in water,” Anal. Chim. Acta 317, 287-293 (1995).
2. Suffren, Y., Rollet, F. G., Reber, C., “Raman spectroscopy of transition metal complexes: molecular vibrational frequencies, phase transitions, isomers, and electronic structure,” Comments on Inorganic Chemistry 32(5-6), 246-276 (2011).
3. Dolenko, T.A., Churina, I.V., Fadeev, V. V., and Glushkov, S. M., Valence band of liquid water Raman scattering: some peculiarities and applications in the diagnostics of water media, J. Raman Spectroscopy, 2000, vol.31, pp. 863-870.
4. Sarmanova, O., Burikov, S., Dolenko, S., Isaev, I., Laptinskiy, K., Neeraj, P., Didem, S. K., Rosenholm, J. M., Shenderova, O., Dolenko, T., “A method for optical imaging and monitoring of the excretion of fluorescent nanocomposites from the body using artificial neural networks,” Nanomedicine: Nanotechnology, Biology, and Medicine 14(4), 1371–1380 (2018).
5. Ali, I., Brunner, H. & Tang, T. Spectral methods for pantograph-type differential and integral equations with multiple delays. Front. Math. China 4, 49–61 (2009). https://doi.org/10.1007/s11464-009-0010-z
6. Esbensen, K.H., Guyot, D., Westad, F., and Houmoller, L.P. Multivariate Data Analysis-In Practice: An Introduction to Multivariate Data Analysis and Experimental Design, 5th ed., CAMO Software AS, 2006.
7. Artursson, T., Eklöv, T., Lundström, I., Mårtensson, P., Sjöström, M. and Holmberg, M. (2000), Drift correction for gas sensors using multivariate methods. J. Chemometrics, 14: 711-723. doi:10.1002/1099-128X(200009/12)14:5/63.0.CO;2-4
8. Gutierrez-Osuna, R. (2000). “Drift reduction for metal-oxide sensor arrays using canonical correlation regression and partial least squares,” in Proceedings of the 7th International Symposium On Olfaction & Electronic Nose, 1–7.
9. Goodall, C. (1991), Procrustes Methods in the Statistical Analysis of Shape. Journal of the Royal Statistical Society: Series B (Methodological), 53: 285-321. doi:10.1111/j.2517-6161.1991.tb01825.x
10. B. Geng, D. Tao and C. Xu, "DAML: Domain Adaptation Metric Learning," in IEEE Transactions on Image Processing, vol. 20, no. 10, pp. 2980-2989, Oct. 2011, doi: 10.1109/TIP.2011.2134107.
11. Efitorov A., Dolenko T., Burikov S., Laptinskiy K., Dolenko S. (2016) Neural Network Solution of an Inverse Problem in Raman Spectroscopy of Multi-component Solutions of Inorganic Salts. In: Samsonovich A., Klimov V., Rybina G. (eds) Biologically Inspired Cognitive Architectures (BICA) for Young Scientists. Advances in Intelligent Systems and Computing, vol 449. Springer, Cham. https://doi.org/10.1007/978-3-319-32554-5_35

Speaker

Trifonov Nikita
Lomonosov Moscow State University
Russia

Discussion

Ask question