(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: demonstrate the feasibility of accelerating Monte Carlo (MC) radiation dose calculations, using deep-learning-based algorithms.
Methods: have developed a convolutional neural network (CNN), called Monte Carlo Denoising Net (MCDNet), which is trained to directly predict the high-photon (low-noise) dose maps from the low-photon (high-noise) dose maps. Two phantoms-based training datasets were used. One training dataset is a unique set of full-body anatomically realistic adult voxel phantoms of various sizes which is used for for patient CT doses. The other training dataset is thirty patients with postoperative rectal cancer who accepted intensity-modulated radiation therapy (IMRT) which is used patient radiotherapy dosimetry. We use 3D Gamma Index Passing Rate (GIPR) to evaluate the performance of predicted dose maps.
Results: the GIPR as the standard, for the CT scan, MCDNet is found to have the ability of predicting dose maps that are equivalent of 9.9×107 photons from corresponding high-noise dose maps of 1.3×106 photons, yielding 76× speed-up in terms of photon numbers used in the MC simulations. For the radiotherapy, the MCDNet can improve the GIPR of dose maps equivalent of 1×107 photons over that of 1×108 photons, yielding over 10× speed-up.
Conclusion: the cases tested, MCDNet has been found to speed-up MC radiation transport simulations involving 3D and heterogeneous patient anatomies for x-ray CT and radiotherapy.
Keywords
Monte Carlo, Dose, Image Processing
Taxonomy
TH- External Beam- Photons: Computational dosimetry engines- Monte Carlo
Contact Email
