Determination of the mass attenuation coefficient over a broad energy range based on the Rayleigh to Compton scattering ratio: Curve-fitting and machine learning methods

In this study, we present two novel methods for determining the mass attenuation coefficient over a broad energy range based on the Rayleigh and Compton scattering spectra collected by a Si(Li) detector. These methods involve developing either curve-fitting models or artificial neural network models to obtain (i) the relationship between the effective atomic number and the Rayleigh to Compton scattering ratio, and (ii) the dependence of the mass attenuation coefficient on both the effective atomic number and photon energy. Using these models, the mass attenuation coefficient can be calculated from input data consisting of the Rayleigh to Compton scattering ratio and photon energy. Next, we applied the proposed methods to 12 different materials, thereby confirming their accuracy. The results show that both methods offer equivalent accuracy. Specifically, within the energy range of 200 keV to 10 MeV, the mass attenuation coefficients determined by these methods exhibit relative deviations of less than 6 % when compared to XCOM (NIST) values. However, in the low-energy range of 50 keV–200 keV, the relative deviations can be significant for multi-element materials (reaching up to 43 %). Overall, these findings demonstrate the feasibility of our proposed methods as additional solutions for determining the mass attenuation coefficient of materials. Two notable advantages of these methods are that they can derive the mass attenuation coefficient across a broad energy range without requiring any information about the sample composition and that the data set used to fit the calibration curves or train the artificial neural network is generated via Monte Carlo simulations.