ESTRO 37 Abstract book

S347

ESTRO 37

2 CancerCare Manitoba, Medical Physics, Winnipeg, Canada 3 CancerCare Manitoba, Medical Physics, WIniipeg, Canada

Radiation Therapy (SBRT), each delivered with dual arc volumetric modulated arc therapy (VMAT). For twelve study patients a dual “MOSkin” MOSFET detector was attached to the anterior surface of the Rectafix during their SBRT treatment. Time resolved measurements were collected by reading out the detectors at a rate of 1 Hz throughout the dose delivery. The measured dose was compared to the planned dose exported from the treatment planning system using the detector position of the day, determined with Cone Beam Computed Tomography (CBCT) imaging. The average difference between the measured and the planned doses over the whole course of treatment for all arcs measured was 9.7% with a standard deviation of 3.6%. The cumulative MOSkin reading was lower than the total planned dose for 64% of the arcs measured. The average difference between the final measured and final planned doses for all arcs measured was 3.4% of the final planned dose, with a standard deviation of 10.3%. The study proved the feasibility of real time measurements of rectal dose during SBRT VMAT treatment of the prostate. It confirmed the difficulty of measurements in a high dose gradient. The effort involved and the found dose differences highlight the importance of a clear clinical indication for any in vivo dosimetry and the need for action limits. As used here, in vivo measurements can be a useful additional safety measure and dose confirmation when introducing a new treatment technique. SP-0663 Scintillation detectors for in vivo dose validation in brachytherapy K. Tanderup 1 1 Aarhus University Hospital, Department of Oncology, Aarhus C, Denmark Abstract text Treatment verification is of specific importance in brachytherapy due to hypofractionation and steep dose gradients. Furthermore, the brachytherapy workflow includes many manual processes, which increases the risk of radiation events, as most misadministrations and near misses are related to human errors. As likelihood and consequences of errors are considered more prominent in brachytherapy than in EBRT, it is a paradox that treatment verification technologies are less developed in brachytherapy. Preliminary results from surveys within GEC-ESTRO activities indicate that less than 10% of clinics perform in vivo dosimetry, while the majority of clinics are interested, if a relevant system had been available. Novel detector technology is currently opening the field of real-time in vivo dosimetry, which can improve the capacity of error detection significantly. Innovations in in vivo dosimetry are currently driven by application of real- time source tracking during treatment delivery through dose rate measurements combined with error detection algorithms. Real-time in vivo dosimetry requires dose rate detectors and currently two promising concepts are being explored: 1) small point detectors in close proximity to the source and 2) EPID panels. Many types of point detectors are available, but the interest in scintillating crystals is growing with the request for small detectors. A number of detector materials are being explored and key characteristics such as reproducibility, sensitivity, energy dependence, and stability over time are being mapped out in the search of detectors which can make in vivo dosimetry most sensitive to errors and which can function optimally in the clinical workflow. SP-0664 EPID-based 3D in vivo dosimetry for SBRT lung and spine – Values and Challenges E. Van Uytven 1 , P. McCowan 2 , T. VanBeek 3 , B.M. McCurdy 2 1 University of Manitoba, Dept of medical physics- Dept of physics and Astronomy, Winnipeg- Manitoba, Canada

Abstract text Introduction

In 2014, CancerCare Manitoba (CCMB) began performing 3D in vivo dose verification of for all fractions of stereotactic body radiation treatments (SBRTs). Findings over a 31-month period were analyzed in detail and reported recently in the Red Journal [1] and will be discussed. The values and challenges associated with clinical implementation of an EPID-based in vivo dosimetry program and on-going sustainable operation will also be discussed. While pre-treatment quality assurance (QA) systems provide verification of linac deliverability and output, in vivo dose verification methods can provide verification of treatment efficacy to the patient. Methods & Materials 100 lung and 15 spine treatments provided a total of 602 fractions of measured data. During treatment delivery, transmission aS1000 EPID images were acquired in cine mode. The EPID-derived fluence maps were used to reconstruct the 3D dose to the planning CT data set. The 3D dose for each fraction was compared to the planned 3D dose distribution (Eclipse treatment planning system). Our measured 3D doses were calculated using an in- house, MATLAB-coded, superposition-convolution collapsed cone convolution (CCC) algorithm which has been previously verified for IMRT and VMAT. Frame average optimization was also performed partway through this study as it was found to substantially increase accuracy in the 3D dose reconstruction. Dose distributions were compared using a 3D γ-test, with a 3%/3mm criteria, on the planning target volume (PTV). We chose pass-rate action levels of 85% for lung and 80% for spine as a starting point for this work. Fractions that failed this action level were analyzed offline by initially comparing the CBCT vs CT as well as treatment parameters and patient setup. Errors were categorized into: EPID-specific, patient setup, anatomical, dose model differences, or unknown. Results For the 537 lung fractions, 126 (23.5%) flagged our action level pass-rate. Of those, the error categories were identified as 82 (65.1%) EPID-specific, 25 (19.8%) patient setup, 10 (7.9%) anatomical, 8 (6.3%) model difference, and 1 (0.8%) unknown. Specifically, of the 82 EPID- specific errors, 53 were acquisition failures (i.e. forgot to schedule imaging, deploy the EPID, computer system crashes, etc.) and 29 were due to frame-averaging artefacts. After optimizing the frame averaging, and assuming we remove human error contributions through increased education, the total flagged fractions would then be 44 (126-82), and the leading category flagging our alert level would be patient-specific, i.e., 35 of 44 fractions or 79.5%. Model differences then make up most of the remaining issues (18.2%). This is an important finding because patient-specific errors are not detectable using standard pre-treatment quality assurance approaches. For the 55 acquired spine fractions, 27 (49%) flagged our action level pass-rate. Of those, 20 were categorized as EPID-specific and 7 were anatomical. Like the above argument, 100% of the errors are patient- specific once the EPID-specific errors are removed.