Parts | Days | Selection | Search | Updates | Downloads | Help
ST: Fachverband Strahlen- und Medizinphysik
ST 3: Artificial Intelligence in Medicine
ST 3.4: Talk
Tuesday, March 22, 2022, 11:45–12:00, ST-H4
A step towards treatment planning for microbeam radiation therapy: fast dose predictions with generative adversarial networks — •Florian Mentzel1, Micah Barnes2, Kevin Kröninger1, Michael Lerch2, Olaf Nackenhorst1, Jason Paino2, Anatoly Rosenfeld2, Ayu Saraswati3, Ah Chung Tsoi3, Jens Weingarten1, Markus Hagenbuchner3, and Susanna Guatelli2 — 1TU Dortmund University, Department of Physics — 2Centre for Medical Radiation Physics, University of Wollongong, Australia — 3School of Computing and Information Technology, University of Wollongong, Australia
Microbeam radiation therapy is a novel and currently pre-clinical radiotherapy treatment based on planar arrays of high intensity sub-millimetre synchrotron gamma rays. Due to good healthy tissue sparing it is a promising candidate e. g. for the treatment of glioblastoma. The dose computations required to plan treatments are currently performed using time-consuming Monte Carlo (MC) simulations with Geant4. The dose computations are complex as steep dose gradients near the beams require very high spatial resolution while the need to take into account the effect of stray radiation over large distances makes small voxel sizes infeasible.
A two-fold approach to these problems is explored: first, a novel data taking method for MC simulations is presented. The method considers both high resolution and long-range effects of stray radiation. Second, a deep learning model based on 3D-UNet GANs is created to mimic dose simulations of Geant4, allowing for very short prediction times.