Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
ST: Fachverband Strahlen- und Medizinphysik
ST 3: Artificial Intelligence in Medical Physics
ST 3.4: Vortrag
Dienstag, 16. März 2021, 14:45–15:00, STa
Tracking of low-energy protons in a proton CT system using machine learning methods — •Miriam Schwarze, Marius Hötting, Kevin Kröninger, Florian Mentzel, Olaf Nackenhorst, and Jens Weingarten — Technische Universität Dortmund, Lehrstuhl für Experimentelle Physik IV
An essential component of a proton CT system is the tracking system for estimating the proton trajectory. Algorithms for reconstructing the proton path through a test object are based on knowledge of the position and flight direction of the proton in front of and behind the object. This information is obtained by detecting the protons in multiple layers of pixel sensors. Using the simulation software Allpix2, the interactions of the protons with the materials and the processes in the pixel sensor are simulated in detail and a data set is created.
The sensors provide an accumulation of hits in each detector layer, which have to be combined to individual tracks. Deep neural networks represent a possible form of pattern recognition. The geometric information of the hits is processed into a three-dimensional image, which serves as input for a Convolutional Neural Network. It is investigated to what extent the Convolutional Neural Network is able to predict the track parameters based on the geometric information and what influence the track density has on the result.