Dortmund 2021 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
ST: Fachverband Strahlen- und Medizinphysik
ST 3: Artificial Intelligence in Medical Physics
ST 3.5: Talk
Tuesday, March 16, 2021, 15:00–15:15, STa
Improving information extracted from glow curves of thermoluminescent personal dosemeters using CNNs — •Evelin Derugin1, Florian Mentzel1, Jens Weingarten1, Jörg Walbersloh2, and Kevin Kröninger1 — 1TU Dortmund, Lehrstuhl für Experimentelle Physik IV — 2Materialprüfungsamt NRW
Personal dose monitoring is essential for a successful radiation protection program for occupationally exposed persons. Thermoluminescence detectors are among the most frequently used dosimeters. The Lehrstuhl für Experimentelle Physik IV at TU Dortmund, in cooperation with the Materialprüfungsamt NRW, is developing multivariate techniques for glow curve analysis using convolutional neural networks (CNNs), to gain additional information about the irradiation scenario, i.e. the number of irradiation fractions or the time of irradiation. These can help to better track the reason for radiation exposure and thereby improve the existing radiation protection concept.
Before a convolutional neural network can be used to predict parameters of a new measurement, large data sets are required for training. Our investigations are based on several thousand measured LiF:Mg,Ti glow curves, which are used for the training of the CNNs.
In this talk, we will present results obtained using the CNNs for a multivariate analysis of the glow curves including information about the performance and the optimization of the neural network.