ESTRO 38 Abstract book

S45 ESTRO 38

show that our model recovers D 95 of 61.1 Gy from the noisy MC input of 54.3 Gy whereas the low noise MC (reference) offers 61.5 Gy. We observe higher Peak signal-to-noise ratio (PSNR) for reference vs denoised (39.70 dB) than reference vs input (28.12 dB) with an improvement factor of 11.58 dB. Moreover, the inference time of our model is only 7s for a dose distribution of an average dimension of 158x512x512. Conclusion We propose an end-to-end, fast and fully automated UNet based framework for denoising the MC dose distributions. The proposed framework offers good generalization ability as it involves no pre-processing and can be trained on any tumor site. It provides comparable dose-volume histogram (DVH) to the MC simulation using 1e⁹ particles and thus, identical D 95 . We obtain a significant reduction in computational time: 7s vs 100 min (MC simulation using 1e⁹ particles).

To capture full range of phantom motion, 3D CT scans of the phantom at end-of-inhale (0 cm) and end-of-exhale (3 cm) were acquired using a helical CT scanner. Static 3x3 cm 2 and VMAT plans were created on the end-of-inhale CT scans of the phantom in Monaco TPS V.5.11.01 to deliver 100 cGy to the center of the tumor. A previously validated BEAMnrc model of our 6 MV Elekta Agility linac was used for all simulations 3 . DOSXYZnrc and 4DdefDOSXYZnrc 3 user codes were used, for stationary and moving anatomy dose simulations, respectively. We used 8´10 7 histories to achieve a statistical uncertainty of 0.8% on a dose grid resolution of 2.0x2.0x2.0 mm 3 . Data from the linac delivery log files were extracted to generate input files for simulations. For 4D simulations, deformation vectors were obtained by deformably registering 4DCT scans of the end-of-exhale to end-of- inhale states using Velocity AI 3.2.0. Deformation vectors, along with the phantom motion trace measured with RADPOS, were used to model the phantom motion. The exact same motion as during irradiations was used in simulations by synchronizing the start of the phantom motion with the linac beam-on time. Results Dose values from MC simulations and measurements at the center of the tumor and bottom surface of the plug were found to be within 1s of experimental uncertainties (2.4%). Agreements on the top surface of the plug (high dose gradient region) were found to be better than 4.0%. On the stationary phantom all dose points from simulations passed a 2%/2 mm gamma analysis [Fig.2 top]. On the moving phantom, passing rates were better than 97.0% [Fig.2 bottom].

OC-0090 Use of a realistic breathing lung phantom to verify 4D Monte Carlo dose calculations S. Gholampourkashi 1 , J. Cygler 1,2,3 , B. Lavigne 2 , E. Heath 1 1 Carleton University, Physics, Ottawa, Canada; 2 The Ottawa Hospital Cancer Center, Medical Phyiscs, Ottawa, Canada; 3 University of Ottawa, Radiology, Ottawa, Canada Purpose or Objective To validate 4D Monte Carlo (MC) simulations of dose delivery to a programmable deformable lung phantom using realistic respiratory motion traces. Material and Methods A previously developed tissue-equivalent deformable lung phantom 1 , was modified to enable realistic breathing motion [Fig.1a]. A piston, attached to a programmable motor, provides variable peak-to-peak (P-P) amplitudes in the sup-inf direction. Measurements were performed on an Elekta Agility linac with the phantom in stationary and breathing states [Fig.1b]. Dose within the tumor, located in the lung, was measured using the calibrated Gafchromic EBT3 film and the RADPOS 4D dosimetry system 2 [Fig.1c]. To measure the dose inside the lung, two RADPOS detectors were mounted on the top and bottom surfaces of the plug [Fig.1d]. RADPOS position tracker recorded the phantom motion with a temporal resolution of a 100 ms.