ESTRO 38 Abstract book

S966 ESTRO 38

however, larger discrepancies were observed in areas of high dose gradients. For center position, score values were (2%/2mm: Th5%, Th20%): 93,73±6,95%, 96,66±5,00%; (3%/3mm: Th5%, Th20%): 98,57±2,04%, 99,14±1,89%, respectively. For others film’s position, the results were (2%/2mm: Th5%, Th20%): 87,51±11,92%, 91,88±10,37%; (3%/3mm: Th5%, Th20%): 96,43±5,78%, 95,79±7,83%. The results presented above, confirm the usefulness of such specially design phantoms for more detailed dosimetric verifications. Conclusion The 3D-phantom verification allows compact dosimetry with simultaneous measurements (saving time) in several planes and different than coronal projection. It’s especially important for complicated intracranial cases (proximity to OARs) and multi-metastasis cases. EP-1786 Towards real-time Monte Carlo dose computation: muscle or brain? M. Alber 1 , N. Saito 1 , M. Söhn 2 1 Heidelberg University Hospital, Department of Radiation Oncology, Heidelberg, Germany ; 2 Scientific RT, GmbH, Munich, Germany Purpose or Objective Real-time Monte Carlo (MC) dose computation will soon be an essential pre-requisite for quality assurance of online- adapted treatment plans. Due to a surge in computation power by parallelization both in GPUs (muscle) and CPUs (brain), this goal is in reach. However, muscle and brain need very specific code optimization for full performance. Realistic benchmarks must be obtained with complete source and collimator models to avoid severe bias in timings. Here, we present how the capacity of CPUs to execute complex code can be harnessed by variance reduction (VR) techniques for high-accuracy MC calculations. Material and Methods The SciMoCa code combines elements of the VMC/XVMC/VMC++ and EGSnrc family of codes with a 4D generic MLC model and a five-element virtual source model. Being optimized for CPU, it employs VR such as history repetition, particle splitting and their counterpart, Russian roulette. Source and collimator models can be much accelerated by VR, however, strong use of VR leads to widely varying particle weights, which slows down dose convergence. Ideally, all elementary energy deposition events (tallies) have the same magnitude. SciMoCa reduces the width of the tally distribution by dynamically adjusting the VR methods on the fly, from the particle source to full absorption. The code is further optimized for memory access, especially by an optimized representation of ICRU-derived cross-section tables. Results In benchmarks against EGSnrc, SciMoCa agrees within statistical uncertainty, with a maximum error of 1.8% recorded for a 6 MeV pencil beam in a lung phantom. A density uncertainty of 1% in this geometry would produce a similar error magnitude. SciMoCa was experimentally validated for Elekta VersaHD, Varian TrueBeam and Accuray Cyberknife accelerators. Typical dose errors in commissioning conditions are in the order of 0.5%. Depending on plan and linac complexity, simulation of the source and collimators accounts for 3-20% of overall computation time, an acceleration by a factor 2-40 by dynamic VR tuning. Timing benchmarks show that SciMoCa is 2.5 times faster than XVMC. More importantly, performance scales almost linearly with logical CPU count (10% performance drop between 4 and 96 cores). This allows clinical dose computations with 1% uncertainty on 2 mm grid size in typically 10-60 seconds on contemporary entry-level server hardware (ca. 24 cores). Conclusion Full source and collimator simulation increases the complexity of MC to an extent that puts GPUs at a

Conclusion Ratio tests performed for different beam qualities leads to range of HU values (150HU-250HU) which should be used to define ArcCheck phantom in TPS during preparation of pre-treatment verification plan. Choosing single HU value despite of beam quality can cause false negative result of pre-treatment verification if local gamma analysis is used. Special care should be taken while using ArcCheck for pre-treatment verification of multienergetic plans. EP-1785 Dosimetric verification of stereotactic treatment plans using 3D-printed phantom and GafchromicEBT3 M. Kruszyna 1,2 , A. Skrobala 1,2 , B. Pawalowski 1 , H. Szweda 1 1 Greater Poland Cancer Centre, Medical Physics Department, Poznan, Poland ; 2 Poznan University of Medical Science, Department of Electroradiology, Poznan, Poland Purpose or Objective The verification of stereotactic plans is usually performed with 2D-array of detectors, in the coronal projection. However, for complex cases, where localization of targets is proximity to organs at risk, there is necessity to use detectors with higher resolution and verification in other projections. The aim of this work was to design and build the 3D-printed phantom for verification with films dosimetry and evaluate the correctness of CyberKnife (CK) stereotactic plans’ realization. Material and Methods 3D-printed phantom (PLA material) it is open cube (10x10x10cm3), filled with water, with the possibility to insert films in two projections: transversal and sagittal. The phantom was used in measurements on CK. The set of 3 to 5 sheets of Gafchromic EBT3 (Ashland) films were simultaneously placed and irradiated in the phantom, the films was placed at the center of target volume (reference position) and at distance of 1 to 2 cm from the reference position. A group of 10 intracranial cases with target’s localization close to optical path (5 pts.) and close to brain stem (5 pts.) were evaluated. According to our clinical workflow all plans were verified and passed gamma criteria L2%/2mm, Th5% using the 2D-array SRS1000 (PTW, Freiburg) with mean result of 97,98%, additionally all were measured in 3D-printed phantom. Analysis were done in OmniProI’mRT software (IBA Dosimetry, v1.6) using gamma evaluation method with criteria: 2%/2mm, 3%/3mm and threshold 5% and 20% of maximum dose, respectively. Results The results for most distributions of planned doses showed high compliance with the obtained measurement data,