Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 11: AI Topical Day – AI in Medicine (joint session ST/AKPIK)
AKPIK 11.2: Vortrag
Donnerstag, 23. März 2023, 14:15–14:30, ZEU/0146
Position reconstruction in proton therapy with proton radiography and machine learning — •Jolina Zillner, Carsten Burgard, Jana Hohmann, Kevin Kröninger, Florian Mentzel, Olaf Nackenhorst, Isabelle Schilling, Hendrik Speiser, and Jens Weingarten — TU Dortmund University, Department of Physics, Germany
In proton therapy precise patient positioning is essential for treatment quality. Current research in proton radiography (pRad) enables imaging of the patient immediately prior to irradiation. The idea is to use such pRad images to verify the patients position.
Therefore a 3D Convolutional Neural Network will be developed in order to predict pRad images depending on the CT image of an object and different translations and orientations. A minimization algorithm can then find the translation and rotation vector for which the predicted image has the smallest difference to a measured pRad image of the object, which can be used to correct the objects position. To predict pRad images, the CNN needs to be trained with pRad images and their related object translation and rotation and the CT-image.
This talk introduces the simulation used to generate these pRad training data. Simulations and reference measurements are performed with a primitive elbow phantom: a 3D-printed 3x3x3 cm3 cube with a T-cavity for gypsum-inlays representing a stretched or bent elbow. The target is implemented in GEANT4 based on CT-data.