Lux Ex Machina – an application of machine learning in glow curve deconvolution

个人剂量测定是辐射防护的一个重要方面，特别是在放射学或核医学等医学领域。这也是一个相关的研究领域，因为目前的剂量监测技术，特别是胶片标记剂量计，已经达到了极限。可行的替代方案例如基于众所周知的热释光（TL）现象。由Materialprüfungsamt NRW、多特蒙德和TU多特蒙德大学联合开发的TL-DOS剂量测定系统使用LiF：Mg，Ti薄层作为热释光材料和快速加热程序。该系统可以通过计算加热过程中发射的光子数和适当的校准来可靠地估计辐照剂量。这里提出的项目的目的是通过模拟基本TL过程并使用现代强大的机器学习算法，最大限度地利用从时间分辨光子信号（辉光曲线）获得的信息。所争取的信息是辐照剂量、辐照和读出之间的时间、一个工作周期内的辐照次数以及剂量计所见的辐射类型。根据模拟数据和从明确定义的光子和中子辐照活动中获得的数据，将对机器学习算法进行训练、测试和验证，并将探索辉光曲线解卷积的全部潜力。

Personal dosimetry is an important aspect of radiation protection, in particular in medical fields such as radiology or nuclear medicine. It is also a relevant field of research as current technologies for dose monitoring, in particular film badge dosimeters, are pushed to their limits. Viable alternatives are e.g. based on the well-known phenomenon of thermoluminescence (TL). The TL-DOS dosimetry system, jointly developed by the Materialprüfungsamt NRW, Dortmund, and the TU Dortmund University, uses a thin-layer of LiF:Mg,Ti as thermoluminescent material and fast heating procedures. The system can be used to reliably estimate irradiation doses by counting the number of photons emitted during the heating process and a proper calibration.The aim of the project proposed here is to maximise the information obtained from the time-resolved photon signal, the glow curve, by simulating the fundamental TL processes and by using modern and powerful machine learning algorithms. The information strived for is the irradiation dose, the time between irradiation and read-out, the number of irradiations within one duty cycle and the type of radiation seen by the dosimeter. Based on simulation data and data obtained from well-defined irradiation campaigns with photons and neutrons, the machine learning algorithms will be trained, tested and validated, and the full potential of glow curve deconvolution will be explored.