Abstract Book

S1017

ESTRO 37

Material and Methods A pre-clinical release of version 15.5 of Varian Eclipse treatment planning system was used to perform both dose calculation accuracy validation as well as a timing study between GPU based calculation and traditional CPU based calculation. The test system had access to three Intel Xeon CPU processors as well as three Tesla P100 GPU cards. Validation was performed using both the AcurosXB and Fourier Transform Dose Calculation (FTDC) calculation algorithms, as used for final dose calculation and IMRT optimisation respectively. A sample set of 100 clinically acceptable treatment plans were used for the testing from a range of anatomical sites including brain, head and neck, breast, lung, mediastinum, oesophagus, and pelvis. The sample included a mix of 3DCRT, IMRT, and VMAT plans for both 6MV and 10MV photons. Dosimetric comparisons were conducted using a standard set of dosimetry criteria per anatomical site as recommended by EviQ (www.EviQ.org.au), as well as direct dose distribution comparison within the Eclipse TPS. Timing comparisons were conducted by comparing dose calculation times under controlled conditions. For the AcurosXB dose calculation, both ‘plan’ and ‘field’ modes were compared. Results When comparing the dosimetric results of CPU versus GPU, there were no differences in dosimetry found across all test cases. Although this is expected, given the complexity around allocation of computing tasks within the GPU architecture, must be validated and confirmed. Small differences of up to 4% local dose (majority less than 1.1 % local dose) were seen when comparing ‘plan’ versus ‘field’ modes, which is direct result of the calculation time optimisation mechanisms used by AcurosXB and has been well documented in literature. These differences were only found in regions very far outside the field and aren’t of significance. When comparing CPU to GPU, the GPU calculation was, on average, 4.4 times faster than CPU with calculation times ranging from 36-533 seconds for CPU and 10-70 seconds for GPU. The fastest calculations were also found to be for the ‘field’ mode calculation for either CPU or GPU. For the FTDC algorithm, GPU was found to be up to 3 times faster compared with CPU.

EP-1880 Auto-planning versus human-driven plan in Hodgkin lymphoma radiation treatment: the OARs perspective S. Clemente 1 , C. Oliviero 1 , V. D'Avino 2 , R. Liuzzi 2 , G. Palma 2 , M. Conson 3 , R. Pacelli 3 , L. Cella 2 1 Azienda Ospedaliera Universitaria "Federico II", Diagnostica Morfologica e Funzionale- Radioterapia- Medicina Legale, Naples, Italy 2 Institute of Biostructure and Bioimaging-CNR, National Research Council of Italy, Napoli, Italy 3 Federico II University School of Medicine, Department of Advanced Biomedical Sciences, Naples, Italy Purpose or Objective Technological advances in Hodgkin lymphoma (HL) radiation therapy (RT) by high conformal treatments potentially increase control over organs-at-risk (OARs) dose distribution. Dose-volume histogram (DVH) predictors and Normal Tissue Complication Probability (NTCP) models in HL patient population have been reported for late side effects such as radiation pneumonitis (RP), hypothyroidism (HT), and cardiovascular diseases (CD), supporting the planning optimization procedures so as to limit OAR complication. However, plan optimization remains a very time- consuming task with great output and operator dependent variability. Purpose of the present study is to report on the evaluation of Pinnacle 3 Auto-Planning (AP) algorithm on female HL cases. Material and Methods Planning CT-scan of 8 female patients with supradiaphragmatic HL in standard supine position were considered. Involved site clinical target volume (CTV) included the upper mediastinal and supraclavear nodes. Planning Target Volume (PTV) was obtained by CTV uniform 10-mm expansion. The following OARs were contoured: bilateral lungs, heart, left anterior descending artery, esophagus, spinal cord, breasts and thyroid. A total dose of 30 Gy was prescribed in 20 fractions. A 'butterfly” (BF) volumetric modulated arc therapy was planned with Pinnacle 3 v. 9.10 (Philips Fitchburg, WI, USA) using SmartArc module and Collapsed Cone Convolution Superposition algorithm dose calculation for a Varian True Beam STx linac. Human-driven (Manual-BF) and AP-BF optimization plans were generated using published priority and constraints on the OARs. Of note, in addition to BF technique, the AP engine was applied to a 2 coplanar disjointed arches (AP-ARC) technique. A single AP optimization list of PTV/OAR clinical goals was created for each patient. Plans were calculated with iso- tumor-coverage criteria. For plan comparison, DVHs of PTVs and OARs were extracted, homogeneity and conformity indices (HI and CI) computed and OARs dose- volume constraints and NTCP models for RP, HT, and CD evaluated. Non-parametric Friedman and Dunn tests were used to identify significant differences between groups. Results AP (AP-BF or AP-ARC) offers comparable coverage of the PTV as the manual plan and a consistently better sparing of OARs (Table).

Conclusion GPU technology has been successfully validated for the first release of GPU ready Eclipse TPS (version 15.5), with no dosimetric differences found between GPU and traditional CPU based calculations. Dose calculation times were significantly faster on GPU, being on average 4.4 times faster than CPU. The fastest calculation times overall were achieved using ‘field’ mode on GPU. IMRT optimisation using FTDC was found to be up to 3 times faster on GPU compared with CPU.