Purpose: Electron backscattering is a phenomenon important in accurate assessment of dose deposition around inhomogeneities, where the spatial energy distribution pattern is significantly perturbed. In the absence of a universal theory, Monte Carlo modeling, serving as the main dose computation tool, is limited to the simulation geometry. An alternative approach of utilizing machine learning modeling offers general predictive power but requires ‘clean’ data. Development of such a dataset is our main purpose.
Methods: For the period 1898 to the present, literature searches were done to find all citable, published references in this field. A set of rules was used for data collection including: (a) only experimental results were included from their original sources, (b) no data was eliminated due to their accuracy, precision, or quality, and (c) for self-consistency all energies, thicknesses, and incident angles were converted in units of keV, nm, and degree, respectively. The database was formatted as a collection of ASCII files for general use.
Results: The present database includes values for 49 elements and 18 compounds across the periodic table for the energy range from 0.1keV to 15MeV. Cu, Al, and Ag are the most commonly studied target materials, accounting for ~25% of all data points on energy-reflection coefficients. Little to no data is available for the majority of elements, while results for complex materials of dosimetric interest are essentially nonexistent. Moreover, only 6% of the measured data were acquired at energies above 100keV.
Conclusion: We developed a comprehensive collection of backscattering coefficients as a function of target atomic number, target thickness, electron energy, and electron incident angle. The database provides a framework for quantitative interpretation of electron backscattering, testing of analytical and Monte Carlo models, and especially for machine learning methods well adapted for the prediction of values missing from the current limited experimental datasets.
Not Applicable / None Entered.