ESTRO 38 Abstract book

S574 ESTRO 38

(beam start-up) should be avoided. Given these constraints, the sensitivity of LOG2DOSE is higher (up to G11) than that of a 2D array (G33). Thus, LOG2DOSE can be a viable QA method for the MRIdian Linac. PO-1034 Development of predictive daily machine quality assurance system to predict forthcoming failures T. Fuangrod 1 , W. Puyati 2 , A. Khawne 2 , M. Barnes 3 , P. Greer 3 1 Chulabhorn Royal Academy, Faculty of Medicine and Public Health- HRH Princess Chulabhorn College of Medical Science, Bangkok, Thailand ; 2 King Mongkut's Institute of Technology Ladkrabang, Department of Computer Engineering, Bangkok, Thailand ; 3 Calvary Mater Hospital Newcastle, Department of Radiation Oncology, Waratah, Australia Purpose or Objective A predictive daily machine quality assurance (QA) system based on machine performance check (MPC) tests was developed. This system can be used for a proactive prevention procedure for estimating forthcoming machine failures. Material and Methods An integration between statistical process control (SPC) and autoregressive integrated moving average (ARIMA) forecast model was applied. Total of 490 daily MPC outputs between 2/2015 – 3/2017 were acquired and each parameter of MPC were arranged, which ordered by testing dates, to be a time series. The time-series optimal ARIMA forecast model and the SPC-based warning level (2 sigma) were calculated from 85% of overall data. The system predicted the one-step ahead of MPC outputs then compared with the warning level. The accuracy of forecast model between predicted and measured MPC tests was evaluated using 15% of overall data. Results Most results of one-step ahead forecasting of each parameter of MPC were measured by mean absolute error (MAE) were less than 0.05, excepted jaw collimation, relative gantry, tangential of kV imager, beam output and uniformity change. Overall average MAE was 0.03 for all MPC tests. The accuracy of predicted MPC tests versus the warning level was 83.83%. The table 1 shows results of accuracy tests for the predictive daily QA system. Table 1. Results of accuracy tests for ARIMA forecast model and accuracy rate of warning stage for each MPC test parameter.

PO-1033 What is required to make logfile-based Monte Carlo a viable QA method for realtime plan adaptation? S. Klüter 1,2,3 , C.K. Spindeldreier 1,2,3 , A. Fees 1,2,3 , G. Major 1,2,3 , J. Debus 1,2,3,4,5,6 , M. Alber 1,2,3 1 Heidelberg University Hospital, Radiation Oncology, Heidelberg, Germany ; 2 National Center for Tumor Diseases NCT, Radiation Oncology, Heidelberg, Germany ; 3 Heidelberg Institute of Radiation Oncology HIRO, Radiation Oncology, Heidelberg, Germany ; 4 German Cancer Consortium DKTK, partner Site Heidelberg, Heidelberg, Germany ; 5 Heidelberg Ion-Beam Therapy Center HIT, Radiation Oncology, Heidelberg, Germany ; 6 German Cancer Research Center DKFZ, Clinical Cooperation Unit Radiation Oncology, Heidelberg, Germany Purpose or Objective Real-time plan adaptation renders common means of treatment delivery QA infeasible. To date, the only available method appears to be linac logfile recording and a high-accuracy secondary dose computation (LOG2DOSE). Here, we explore the level of detail and accuracy required for Monte Carlo (MC) simulation of the MRIdian Linac (ViewRay Inc, USA) as well as the necessary stability and precision of linac and MLC to afford a sufficient level of accuracy and sensitivity for daily QA of real-time adapted treatment plans. Material and Methods In order to detect dose delivery errors with LOG2DOSE with high sensitivity, two prerequisites must be met: 1. the logfile information needs to be precise in all parameters, which relates to MLC properties as well as linac characteristics, 2. the MC model needs to be accurate in every aspect of source and collimator simulation, i.e. it is not acceptable that errors cancel out. From the combination of logfile precision and MC accuracy we estimate the maximum 3D dose errors that would go unnoticed with LOG2DOSE. The data presented here were obtained with a prototype of an MR-compatible motorized water phantom, various MR-compatible ionization chambers (Standard Imaging, PTW), a micro-diamond detector (PTW), and two different ionization chamber (IC) arrays (Sun Nuclear, PTW). A set of specific measurements was designed to investigate the MC model of the ViewRay TPS with respect to photon spectrum, energy fluence distribution and MLC model. Logfile precision was evaluated with respect to mechanical stability of the MLC and its calibration, beam start-up behaviour and MU-linearity. Results The MC source model proved extremly accurate, confirmed by output factors and depth dose curves within 0.5%. Transmission/leakage through single stacks of the double-stack MLC was initially off by 50% (mean) and up to 350% (peak), reducing the leaf width by 0.075 mm fixed this issue. Differences between actual and recorded linac behaviour originated from two sources: during beam start-up, the linac output can be between 1% and 6% less for the first 10 monitorunits (MU), with associated energy-dependent fluctuations of the cross-profile. Further, the leaves show a mean hysteresis of 0.4 mm. Leaf calibration remained stable during 6 months. The 2D array was reliable down to a 3%/3mm gamma (G33) criterion but showed spurious disagreements for more sensitive criteria, caused by coarse spacing and large volume averaging of the ICs in the array. Conclusion Issues with MC originated from an overly idealized linac model. Precise adjustment to reality including manufacturing and calibration tolerances is required. Logfiles contain residual errors that could be considered in MC simulations, but for practical reasons very narrow fields (leaf hysteresis) and less than 10 MU per segment