Advancing liquid scintillation spectra deconvolution for radiological characterization: comparative techniques for...

In order to effectively control the level of radioactivity, faster analysis processing times are required. Therefore, new protocols must be developed and existing ones updated. Mathematical deconvolution methods, such as those based on adjusting the shape of the individual spectra of the isotopes, can be applied to separate these spectra, allowing a faster identification and quantification of the radioisotopes of interest. In this paper, several techniques based on Fourier series, Laguerre polynomials, Legendre polynomials, and Savitzky-Golay filter are used for spectral filtering, while multivariate statistical calibration techniques, such as least squares and partial least squares regressions, are used for spectral deconvolution and the prediction of the activity of beta emitters measured by liquid scintillation counting ( $_{}^{63}$ Ni, $_{}^{55}$ Fe, $_{}^{14}$ C, $_{}^3$ H, $_{}^{89}$ Sr, and $_{}^{90}$ Sr). The results obtained for all the techniques analyzed are accurate, considering the maximum relative biases obtained in comparison with the reference activities and the maximum reconstruction errors of the spectra. Different relative biases are observed in the calculation of the activities of each radioisotope under study. Generally, the relative biases obtained are less than 20% for all techniques applied in most of the test samples. However, this difference is not so evident in terms of the reconstruction of the spectra, obtaining a maximum error of about 4.5%. Among all the techniques applied, the combination of Legendre polynomials or Laguerre polynomials and partial least squares regression offers the best results.