Monte-Carlo-Informed Shape Learning for Gain Shift Correction based on LaBr3:Ce,Sr Intrinsic Activity

Abstract

Translate Article

Coincidences in La-138 are simulated with the Monte-Carlo solver tRAYcy, modelling the intrinsic activity $\mathrm{LaBr3}: \mathrm{Ce}, \mathrm{Sr}$ detector [1] and the influences of additional potassium K-40. Based on this data, a decision tree regressor [2] is trained to estimate a numerical value for contribution of additional K-40, to improve retrieval of the peak position. Lanthanum bromide, typically doped with Cerium or Strontium ($\left.\mathrm{LaBr}_{3}: \mathrm{Ce}, \mathrm{Sr}\right)$ has become a versatile detector material in the high resolution market of scintillators. With an FWHM often better than 3%, the material is a notable attractor for homeland security applications [3]. Due to its intrinsic activity of La-138 [4], this scintillator shows a complex background radiation spectrum containing many features - a fact thoroughly discussed in literature. Some users understand this background radiation as distortion, yet it is also known to be exploitable for automatic calibration against peak shifts. The most prominent feature of the LaBr3:Ce,Sr intrinsic background is a peak structure at between 1435 keV and 1468 keV. It is due to a coincidence of the La-138 gamma line at 1435.7 keV and its series of X-rays emitted between 4 keV and 37 keV. The 1460 keV gamma line of natural K-40 usually overlaps with this LaBr3 structure, even increasing the complexity of the spectroscopic structure. The paper presents a novel technique that decomposes the overlapping to find an accurate gain prediction for stabilisation. The proposed decision tree uses the Gini-Impurity as measure for splitting its nodes and yields reasonable performance on finding the right K-40 amount just based on shape analysis. The stabilisation is further tested with a detector in a laboratory climate chamber during temperature ramps to emulate real-life usage e.g. for nuclear security scenarios.