- Email: [email protected]
First Experience With Real-Time EPID-Based Delivery Verification During IMRT and VMAT Sessions
Accepted Manuscript First experience with real-time EPID based delivery verification during IMRT and VMAT treatments Henry C. Woodruff, PhD, Todsaporn Fuangrod, MS, Eric Van Uytven, PhD, Boyd M.C. McCurdy, PhD, Timothy van Beek, Shashank Bhatia, PhD, Peter B. Greer, PhD PII:
S0360-3016(15)03062-X
DOI:
10.1016/j.ijrobp.2015.07.2271
Reference:
ROB 23078
To appear in:
International Journal of Radiation Oncology • Biology • Physics
Received Date: 6 February 2015 Revised Date:
25 June 2015
Accepted Date: 13 July 2015
Please cite this article as: Woodruff HC, Fuangrod T, Van Uytven E, McCurdy BMC, van Beek T, Bhatia S, Greer PB, First experience with real-time EPID based delivery verification during IMRT and VMAT treatments, International Journal of Radiation Oncology • Biology • Physics (2015), doi: 10.1016/ j.ijrobp.2015.07.2271. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
First experience with real-time EPID based delivery verification during IMRT and VMAT treatments Henry C. Woodruff, PhD1, Todsaporn Fuangrod, MS2, Eric Van Uytven, PhD3, Boyd M.C. McCurdy, PhD3, Timothy van Beek3, Shashank Bhatia, PhD4 and Peter B. Greer,
Faculty of Science and IT, School of Mathematical and Physical Sciences, University of Newcastle,
NSW 2308, Australia 2
Faculty of Engineering and Built Environment, School of Electrical Engineering and Computer
M AN U
Science, University of Newcastle, NSW 2308, Australia 3
SC
1
RI PT
PhD1,4
Division of Medical Physics, CancerCare Manitoba, 675 McDermot Avenue, Winnipeg, Manitoba
R3E 0V9, Canada, Department of Physics and Astronomy, University of Manitoba, Winnipeg, Manitoba R3T 2N2, Canada, and Department of Radiology, University of Manitoba, Winnipeg, Manitoba R3T 2N2, Canada 4
TE D
Department of Radiation Oncology, Calvary Mater Newcastle Hospital, Locked Bag 7, Hunter region Mail Centre, Newcastle, NSW 2310, Australia
EP
Reprint requests to: Henry C. Woodruff, University of Newcastle, School of Mathematics and Physical Sciences, University Drive, Callaghan NSW 2308, AUSTRALIA; Tel: (+61) 432 911 085; E-mail: [email protected]
AC C
Conflict of interest: none.
ACKNOWLEDGEMENTS The authors gratefully acknowledge funding support from the Radiation Oncology Institute Grant# ROI2013-912, Cancer Institute New South Wales Hunter Translational Cancer Research Unit and the Calvary Mater Newcastle Radiation Oncology Research Fund.
ACCEPTED MANUSCRIPT Summary
AC C
EP
TE D
M AN U
SC
RI PT
This prospective clinical trial evaluated whether electronic portal imaging device images, in combination with a series of predicted portal images, can determine in real-time whether external beam radiation treatments are being delivered as planned. The data collected in this study will serve to determine action thresholds for future in-vivo real-time transit dosimetry software. Electronic portal transit image comparisons can simultaneously detect geometric and dosimetric treatment deviation stemming from patient and linear accelerator variations.
1
ACCEPTED MANUSCRIPT
ABSTRACT Purpose: Gantry-mounted megavoltage electronic portal imaging devices (EPIDs) have become ubiquitous on linear accelerators. WatchDog is a novel application of EPIDs, in which the image
RI PT
frames acquired during treatment are used to monitor treatment delivery in real time. We report on the preliminary use of WatchDog in a prospective study of cancer patients undergoing intensity
modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT), and identify the
SC
challenges of clinical adoption.
Methods and Materials: At the time of submission 28 cancer patients (head and neck, pelvis and
M AN U
prostate) undergoing fractionated external beam radiation therapy (24 IMRT, 4 VMAT) had at least one treatment fraction verified in real time (131 fractions or 881 fields). EPID images acquired continuously during treatment were synchronized and compared to model generated transit EPID images within a frame-time (~0.1 s). A χ-comparison was performed to cumulative frames to gauge overall delivery quality, and the resulting pass rates reported graphically during treatment delivery.
analysis.
TE D
Every frame acquired (between 500 and 1500 per fraction) was saved for post processing and
Results: The system reported mean ± 1 standard deviation real-time χ 91.1 ± 11.5% (83.6 ± 13.2%)
EP
for cumulative frame χ analysis with 4%, 4mm (3%, 3mm) criteria, global over the integrated image.
AC C
Conclusions: A real-time EPID based radiation delivery verification system for IMRT and VMAT has been demonstrated that aims to prevent major mistreatments in radiation therapy.
Keywords:
Real-time in-vivo dosimetry, EPID dosimetry, Intensity-modulated radiotherapy
2
ACCEPTED MANUSCRIPT INTRODUCTION External beam radiation therapy (EBRT) has evolved to become a complex delivery method for ionizing radiation. Intensity modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) lead the field of highly conformal delivery methods [1, 2].
RI PT
Due to the complex and non-intuitive fluences, often paired with large dose gradients, accuracy of delivery is usually verified by patient specific pre-treatment dosimetric
measurements [3-7]. However, a recent investigation at the Netherlands Cancer Institute
SC
revealed that nine of 17 serious errors identified in a cohort of 4337 patients would have been missed without in-vivo verification of radiation delivery [8]. Subsequently, in-vivo electronic
M AN U
portal imaging device (EPID) dosimetry for post-treatment analysis has been implemented by a few institutions [9-12]. With the increasing use of stereotactic body radiotherapy with very large doses delivered in 1-5 fractions, post-delivery analysis is not sufficient to detect and correct errors. We have developed and tested with phantom irradiations a real-time cine EPID
TE D
based verification system (WatchDog), enabling detection of gross treatment delivery errors prior to delivery of substantial radiation to the patient [13, 14]. A reference dataset of EPID images is predicted using parameters exported from the treatment planning system [15, 16],
EP
and compared to the stream of EPID images acquired during treatment. The comparison is
AC C
performed within a frame-time (~0.1 sec) and allows for both geometric (collimator positions, gantry angle, patient anatomy and positioning) and dosimetric (dose accuracy as a function of control point (CP) and gantry angle) verifications to be performed. Such a verification system provides an additional tool that can assist with prevention of major mistreatments in radiation therapy [10, 17-21].
In this paper we report on the first clinical demonstration of real-time EPID based dose delivery verification for patients undergoing IMRT and VMAT treatments. At the time of
3
ACCEPTED MANUSCRIPT submission, we have acquired EPID cine-images during treatment for 28 cancer patients (head and neck, pelvis, anal canal and prostate). The aims of this paper are to: 1) investigate accuracy of synchronization of measured and predicted image sets during patient deliveries; 2) determine prediction model accuracy for transit patient images using integrated image
RI PT
assessment; 3) determine initial results for real-time patient treatment verification with the system; 4) determine challenges to clinical adoption of the system.
SC
METHODS AND MATERIALS
M AN U
Overview and patient details
The study received ethical approval by the local human research ethics committee and the first phase was designed to be observational only, therefore no intervention was performed based on the monitoring information. The eligibility criteria were designed to ensure that
TE D
there would be a characteristic sample of treatment fields per patient. Twenty-eight patients with ages ranging from 51 – 78 years, with a median age of 67, had at least one fraction verified in real-time. All plans were hyper-fractionated and sites treated with IMRT were
EP
prostate (17 patients), head and neck (6 patients), and anal canal (1 patient) and four VMAT patients were treated in the pelvis area. The treatments were planned using Eclipse (Varian
AC C
Medical Systems, Palo Alto, CA) v11.
Linear accelerator, MLC, EPID, OBI and acquisition mode Two C-Series Trilogy linacs (Varian Medical Systems, Palo Alto, CA) equipped with Millennium 120-leaf MLCs were used to generate the 6 MV research (open air, phantom) and curative (patient) radiation fields. Megavoltage (MV) EPIDs (Varian PortalVision aS1000 flat panel detectors) were used to acquire the images, the detector square pixel having a side length of 0.0392 cm, yielding a total area of approximately 40 × 30 cm2. All data were
4
ACCEPTED MANUSCRIPT acquired in integrated acquisition mode by the clinical treatment software using the 4D Integrated Treatment Console (4DITC) PC. To capture image frames custom MATLAB/C# (MathWorks, Natick, MA, USA) code was developed in-house with a dual-base framegrabber card (Matrox Solios SOL 2M EV CLB), housed in a separate PC and connected via
RI PT
two camera-link cables to ports on the 4DITC and on-board imager (OBI) computers. The system captures every MV and kV frame, constituting a time-lapse series of individual MV frames and kV source angles acquired during dose delivery at frame rates of 7.455 fps and
SC
10.92 fps, respectively. Gantry angles were derived from kilovoltage (kV) source rotation information encoded in kV frames (the kV source was not active). Synchronization of the
M AN U
EPID image with the corresponding kV image containing the gantry angle information was achieved through time-stamps. All IMRT (VMAT) fields were clinical fields, with 400 (600) monitor units per minute (MU/min) nominal dose-rate respectively, and the EPID was positioned at 150 cm source-to-detector-distance (SDD). All EPID frames were dark-field
Prediction Model
TE D
and flood-field corrected by the 4DITC.
EP
For each treatment field a series of predicted EPID images were calculated at predetermined
AC C
meterset CP intervals (based on dose or gantry angle progression in IMRT or VMAT, respectively) using the method of Chytyk et al. [15, 16], providing a sequence of images for the entire beam delivery, as described in [13] and [14]. The model takes into account the energy response of the EPID by separating the energy-fluence into 15 energy bins, and incorporates support arm backscatter [22] and the attenuation and scatter caused by the patient and treatment couch in the beam, based on planning computed tomography (CT) data. The software automatically removes the CT couch from the CT volume and inserts the desired treatment couch. DICOM-RT files were used to input the treatment plan data into the
5
ACCEPTED MANUSCRIPT prediction model, and DICOM image files were used to import the CT volume for ray tracing. Artefacts in the CT volume introduced by patient implants also introduce artefacts in the predicted images, which were modeled at the rate of two (three) per CP for IMRT (VMAT) treatments to balance the need for a fine grid of predicted images with
RI PT
computational and storage limitations. The model was optimized for parameters such as intraleaf leakage (including tongue and groove effect), MLC leaf offsets, total backscatter and extra-focal contribution. Each predicted frame contains a distribution of grayscale values that
SC
are calibrated using a factor derived from the ratio of model prediction and measured EPID images at central axis for a 10x10 cm2 field at 150 cm SDD and 100 MU. The model was
M AN U
validated against measured EPID images for a wide variety of IMRT and VMAT test fields and clinical plans with and without phantoms present ([13], [14]). Image synchronization
The EPID acquires images at frame-rates set by the manufacturer and the image stream is not
TE D
synchronized to the linac’s dose delivery. Dynamically controlled dose-rate variations are varied “on the-fly” by the linac control systems and cannot be predicted. Therefore a method
EP
is needed to synchronize each measured frame to the corresponding predicted frame. For IMRT deliveries MLC leaf positions extracted from predicted and measured images are used
AC C
for synchronization, as described in [13]. The following changes were implemented in order to improve real-time processing: (i) the collimator angle was extracted from the treatment planning system (TPS) data instead of from the measured images; (ii) rather than rotating each image to 0° collimator angle to match the MLC leaf template, the template was rotated to match the collimator angle; (iii) the search space for predicted images available for synchronization was narrowed by moving boundaries (we selected 2 CPs behind and 10 CPs ahead of the currently synchronized CP). The accuracy of synchronization was assessed by comparing the synchronization index, which assigns each measured frame to the
6
ACCEPTED MANUSCRIPT corresponding predicted frame, with a reference step function (stepped due to the discrete number of CPs). This fiducial is derived after beam delivery from the ratio of the number of frames (“measured”/“predicted”).
RI PT
Due to the gantry-angle dependency of VMAT control systems, synchronization based on gantry-angles was chosen for VMAT deliveries, as described in [14]. The model uses TPS data to calculate EPID images at predetermined gantry angle increments. Measured EPID
M AN U
Real-time comparison and verification
SC
images are synchronized to the predicted images using gantry angle.
The system evaluates a cumulative image where each new measured image is added to the summed image and compared to the cumulative predicted image up to the current synchronization point, without normalization, yielding an evaluation which becomes less responsive to errors as a function of beam-on time but allows for dosimetric as well as
TE D
geometric evaluation of the treatment delivery [13]. Measured frames were binned to ½ resolution (512 x 384 pixels2) and 2-D comparisons were evaluated by using the χ method
EP
[24] to minimize the use of computing resources and perform all calculations within a frame time (~ 0.1s). The main aim of this study was to implement and gain initial experience with
AC C
the system and detect only major deviations from the treatment course, so a global criteria of 4%, 4 mm for pixels above a 10% of maximum signal threshold were used (dose criterion and threshold are calculated on the cumulative measured image), and a treatment field was flagged as failed if more than 4 consecutive frames fell below a χ pass-rate of 40%. These criteria were chosen from analysis of preliminary patient data, which showed large interfraction variation.
7
ACCEPTED MANUSCRIPT Clinical data acquisition The clinical workflow for a patient undergoing real-time verification differs little from a regular patient, with the presence of the extended EPID throughout treatment the only change apparent to the patient. The software has no ability to interfere directly with the treatment
RI PT
delivery. Prior to first fraction delivery treatment plan parameters and patient CT are exported from the TPS and the predicted frame set calculated. The predicted images are automatically encapsulated in a file containing all relevant patient and treatment parameters, which is then
SC
sent to the WatchDog PC at the linac. An integrated imaging template is attached to the
treatment plan to activate EPID image acquisition during delivery. Once the patient has been
M AN U
set-up for treatment, the radiation therapist (RT) selects the patient using the WatchDog graphical user interface (GUI) and presses “start” before the beam is delivered. The system automatically detects when the beam is on and automatically changes to next field when necessary. A simple “red / green light” flag and a progress bar are shown next to each field,
TE D
indicating delivery status. χ pass-rates are also reported graphically and in real-time. Figure 1 shows an example of the initial GUI implemented for patient treatment verification. All acquired clinical images and results were saved and system problems that prevented data
EP
acquisition were recorded and investigated. The real-time comparison produces a χ pass-rate
AC C
value for each measured image during the delivery of an IMRT field or VMAT arc. Here we present these results as the mean and standard deviation of these χ results. For each treatment site, the χ results are grouped for all patients and the mean and standard deviation quantified.
Figure 1: WatchDog graphical user interface for real-time treatment verification. The top left panel shows patient and treatment information, while the bottom left panel shows a progress
8
ACCEPTED MANUSCRIPT bar for each beam, the associated gantry angle, and a red/green status indicator. The plot in the upper right shows the real-time χ pass-rates and is updated with every new measured frame, with the measured EPID image displayed underneath.
RI PT
RESULTS Image synchronization
The accuracy of both IMRT and VMAT synchronization algorithms was previously found to
SC
be within 2 CPs (approx. 1-2% of total dose) when imaging through air or a phantom ([13, 14]). To assess the impact of variation in patient anatomy and/or positioning on
M AN U
synchronization accuracy 20 fields were measured during treatment with a patient present in the beam (10 each from H&N and prostate). These were subsequently synchronized to modeled frames with and without a patient CT scan included in the prediction model. The absence of the patient in the model was used to represent the effect of extreme anatomical
TE D
variation on the synchronization.
Figure 2 shows the χ pass-rates for the cumulative comparison for a prostate field, together
EP
with the associated synchronization indices, fiducial step function and their difference.
Figure 2: Prostate IMRT results for a single field for real-time verification (top),
AC C
synchronization indices (middle) and deviation from optimal synchronization in CPs (bottom) when comparing measured data with a patient to a model with the patient in the beam (left) and without (right).
The mean deviation (± standard deviation) of the synchronization index from the reference function was 0.04 ± 0.68 CPs including the patient compared to -0.07 ± 0.86 CPs without a patient in the model for the 20 patient deliveries. This shows that the synchronization
9
ACCEPTED MANUSCRIPT precision can be assumed to be independent of patient positioning or anatomical variation. The poor cumulative pass-rates for Figure 2(b) are due to the model not including the effect of the patient attenuation and scatter. VMAT deliveries were synchronized using measured
RI PT
and planned gantry angles which are not affected by imaging considerations [14].
Model and patient variability
The prediction model accuracy was validated against integrated patient transit EPID images
SC
acquired during treatments for a variety of IMRT (81 prostate, 43 head and neck and 14
pelvis) and VMAT (8 pelvis) fields for a total of 28 patients. Average ± 1 standard deviation
M AN U
integrated image χ pass-rates using 4%, 4mm (3%, 3mm) criteria, global over the integrated image, were found to be 96.7 ± 14.3% (92.7 ± 16.3)%, ranging from 61.4% to 100% (55.3 to 100%), with the largest divergence from the model found in prostate fields at 94.8 ± 15.2% (91.0 ± 15.9%) and the least divergence found in H&N fields at 98.1± 8.9% (95.6 ± 10.9%).
TE D
Delivery of two IMRT fields to one prostate patient in particular were responsible for all χ
EP
pass-rates < 80% and will be discussed separately as a case study.
Real-time error detection
AC C
The system results for real-time verification were analysed for the 28 patients. Total cumulative lag (i.e. lag at the end of each field) between delivery and reporting of χ pass-rates depends on the size of the delivered field as measured by the EPID (due to longer calculation times for the χ metric for larger fields), and was found to be between 0.3 s for smaller prostate fields and up to 0.9 s for the largest head and neck fields. Average cumulative χ passrates using 4%, 4mm (3%, 3mm) were 91.1% (83.6%) with a standard deviation of 11.5% (13.2%). Figure 3 summarises the data in a boxplot, sorted by treatment site.
10
ACCEPTED MANUSCRIPT Figure 3: Boxplot of real-time cumulative image χ passrates (4%, 4mm left, 3%, 3mm right, global) sorted by treatment site. The bottom and top of the box represent the first and third quartiles, and the separating line the median. The ends of the whiskers represent the
RI PT
minimum and maximum of all of the data for that category.
SC
Case study
A system which compares measured and predicted 2D transit EPID images will be sensitive
M AN U
to a range of dosimetric as well as geometric departures from the planned delivery, including internal anatomical changes. Error! Reference source not found. Mean real-time cumulative χ pass-rates (4%,4mm) for a prostate patient (7 field IMRT) over the course of 44 days (19 fractions) were 76.5% with a mean standard deviation of 6.3, ranging from (50.4-
TE D
94.8)%. No correlation was found between χ pass-rates and positioning metrics (couch parameters, set-up shifts, patient notes, set-up kV image analysis including gold seed and bone structure alignment) or patient weight. A single cone-beam CT was acquired in the last
EP
week of treatment that indicated large sections of the rectum filled with gas in contrast to the planning CT, which showed a rectum full of fecal matter. Moreover, increased organ motion
AC C
cannot be disregarded since prostate motion has been shown to increase in the presence of gas in the rectum (see, e.g. [25]), which could compound the departure from simulated conditions [26].
DISCUSSION A novel method for real-time verification of IMRT/VMAT treatment deliveries using EPID imaging was successfully implemented in a clinical setting in a cohort of 28 patients. Realtime EPID dosimetry complements routine quality assurance and presents a final safety net in
11
ACCEPTED MANUSCRIPT the EBRT process chain. As opposed to recent EPID-based in-vivo dosimetry solutions (discussed in detail in [9] and [10]), our system is designed to detect gross errors as they occur, before delivery of substantial dose to the patient, giving the therapists the option to
RI PT
interrupt a treatment and inquire into the causes of the discrepancy.
There are no precedents or recommendations to follow regarding real-time EPID transit
dosimetry. A global (over the integrated image) χ comparison criteria of 4%, 4 mm for pixels
SC
above a 10% threshold were used, and a treatment field would have been flagged as failed if more than 4 consecutive frames fell below a χ pass-rate of 40% for cumulative frame
M AN U
comparison. No errors, as per these parameters, were detected in the 131 fractions or 881 fields which were verified using the system, although low χ pass-rates prompted further enquiries into one patient (7 field prostate IMRT). A challenge to any image based analysis that distils the difference in dose distributions across two images down to one or two numbers
TE D
is how to distinguish between variations in the patient’s anatomy and changes in the delivered dose. Both cases imply that the dose was not delivered as planned, which is reported by our system, although false positives are possible in cases where large image differences (and
EP
hence low χ pass-rates) caused by, e.g. changes in the bony anatomy, may lead to small
AC C
differences in the dose delivered to the target volume. Optimal χ criteria will be the objective of further studies based on larger patient cohorts. These will be designed to detect clinically relevant errors and not small errors owing to transient anatomical changes [26-28]. The system is not able to verify fields where collisions between EPID and the couch or the patient are possible, such as non-coplanar beams. Patient with implants, which create large artefacts in the planning CT volume and also lead to large artefacts in the predicted EPID images, were excluded from this study
12
ACCEPTED MANUSCRIPT There are several issues for clinical adoption that have been identified in these preliminary patients. The system uses a research computer that interfaces with the clinical imaging systems via a third party frame-grabber card, and there are challenges that are being overcome to make this interface robust. The forward prediction model takes up to 12 hours to
RI PT
produce a stream of predicted EPID images, but these calculations can be performed in
batches and unsupervised, and the software supports dedicated graphical processing units, which can halve prediction times. There is currently no method to change the order of the
SC
fields as they are delivered, a feature that the RTs involved in this study have asked to be introduced in future versions. Image acquisition can also be missed, as the system must be
M AN U
separately activated before treatment although the frequency of this decreased throughout the study. Cone-beam CT (CBCT) acquisition stops the stream of kV dark frames to the framegrabber, presenting a problem specific to VMAT deliveries that require constant update of the gantry angles. The current solution is to add a single kV image setup-field (e.g. anterior-
CONCLUSION
TE D
posterior) in addition to the CBCT, which must be moded up just prior to treatment.
EP
We have presented the successful clinical implementation of WatchDog, an independent system to verify EBRT treatment in real-time using EPIDs. The frame-stream is synchronized
AC C
to the predicted images and evaluated using the χ comparison. Average real-time cumulative χ global pass-rates using 4%, 4mm (3%, 3mm) over the integrated image were 91.1% (83.6%) with a standard deviation of 11.5% (13.2%). A clinical case study involving a patient with gas in the rectum illustrated the system’s ability to indicate deviations from the original plan that would have otherwise remained undetected.
13
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
REFERENCES 1. Yu, C.X., Intensity-modulated arc therapy with dynamic multileaf collimation: an alternative to tomotherapy. Physics in Medicine and Biology, 1995. 40(9): p. 1435. 2. Otto, K., Volumetric modulated arc therapy: IMRT in a single gantry arc. Medical Physics, 2008. 35(1): p. 310-317. 3. Both, S., et al., A study to establish reasonable action limits for patient specific IMRT QA. 2007. Vol. 8. 2007. 4. van Elmpt, W., et al., The next step in patient-specific QA: 3D dose verification of conformal and intensity-modulated RT based on EPID dosimetry and Monte Carlo dose calculations. Radiotherapy and Oncology, 2008. 86(1): p. 86-92. 5. Ling, C.C., et al., Commissioning and Quality Assurance of RapidArc Radiotherapy Delivery System. International Journal of Radiation Oncology*Biology*Physics, 2008. 72(2): p. 575-581. 6. Van Esch, A., et al., Implementing RapidArc into clinical routine: A comprehensive program from machine QA to TPS validation and patient QA. Medical Physics, 2011. 38(9): p. 5146-5166. 7. O'Daniel, J., et al., Volumetric-modulated arc therapy: effective and efficient end-toend patient-specific quality assurance. Int J Radiat Oncol Biol Phys, 2012. 82(5): p. 15671574. 8. Mans, A., et al., Catching errors with in vivo EPID dosimetry. Medical Physics, 2010. 37(6): p. 2638-44. 9. van Elmpt, W., et al., A literature review of electronic portal imaging for radiotherapy dosimetry. Radiotherapy and Oncology, 2008. 88(3): p. 289-309. 10. Mijnheer, B., et al., In vivo dosimetry in external beam radiotherapy. Medical Physics, 2013. 40(7): p. 070903. 11. Berry, S.L., et al., Implementation of EPID transit dosimetry based on a through-air dosimetry algorithm. Medical Physics, 2012. 39(1): p. 87-98. 12. Berry, S.L., et al., Initial clinical experience performing patient treatment verification with an electronic portal imaging device transit dosimeter. Int J Radiat Oncol Biol Phys, 2014. 88(1): p. 204-9. 13. XXXXX 14. XXXXX 15. Chytyk, K. and B.M.C. McCurdy, Comprehensive fluence model for absolute portal dose image prediction. Medical Physics, 2009. 36(4): p. 1389-1398. 16. Chytyk-Praznik, K., et al., Model-based prediction of portal dose images during patient treatment. Medical Physics, 2013. 40(3): p. 031713. 17. Mayles, W.P.M., The Glasgow Incident — a Physicist's Reflections. Clinical oncology (Royal College of Radiologists (Great Britain)), 2007. 19(1): p. 4-7. 18. Williams, M.V., Radiotherapy Near Misses, Incidents and Errors: Radiotherapy Incident at Glasgow. Clinical oncology (Royal College of Radiologists (Great Britain)), 2007. 19(1): p. 1-3. 19. Derreumaux, S., et al., Lessons from recent accidents in radiation therapy in France. Radiat Prot Dosimetry, 2008. 131(1): p. 130-5. 20. Ortiz Lopez, P., et al., ICRP publication 112. A report of preventing accidental exposures from new external beam radiation therapy technologies. Ann ICRP, 2009. 39(4): p. 1-86. 21. Bogdanich, W. Radiation offers new cures, and ways to do harm. The New York Times, 2010. 22. Rowshanfarzad, P., et al., Measurement and modeling of the effect of support arm backscatter on dosimetry with a Varian EPID. Medical Physics, 2010. 37(5): p. 2269-2278.
14
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
23. XXXXX 24. Bakai, A., M. Alber, and F. Nüsslin, A revision of the γ-evaluation concept for the comparison of dose distributions. Physics in Medicine and Biology, 2003. 48(21): p. 3543. 25. Ogino, I., et al., Reduction of Prostate Motion by Removal of Gas in Rectum During Radiotherapy. International Journal of Radiation Oncology*Biology*Physics, 2008. 72(2): p. 456-466. 26. Kroonwijk, M., et al., In vivo dosimetry for prostate cancer patients using an electronic portal imaging device (EPID); demonstration of internal organ motion. Radiotherapy and Oncology, 1998. 49(2): p. 125-132. 27. McDermott, L.N., et al., Replacing Pretreatment Verification With In Vivo EPID Dosimetry for Prostate IMRT. International Journal of Radiation Oncology*Biology*Physics, 2007. 67(5): p. 1568-1577. 28. Persoon, L.C., et al., Interfractional trend analysis of dose differences based on 2D transit portal dosimetry. Phys Med Biol, 2012. 57(20): p. 6445-58.
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
S0360-3016(15)03062-X
DOI:
10.1016/j.ijrobp.2015.07.2271
Reference:
ROB 23078
To appear in:
International Journal of Radiation Oncology • Biology • Physics
Received Date: 6 February 2015 Revised Date:
25 June 2015
Accepted Date: 13 July 2015
Please cite this article as: Woodruff HC, Fuangrod T, Van Uytven E, McCurdy BMC, van Beek T, Bhatia S, Greer PB, First experience with real-time EPID based delivery verification during IMRT and VMAT treatments, International Journal of Radiation Oncology • Biology • Physics (2015), doi: 10.1016/ j.ijrobp.2015.07.2271. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
First experience with real-time EPID based delivery verification during IMRT and VMAT treatments Henry C. Woodruff, PhD1, Todsaporn Fuangrod, MS2, Eric Van Uytven, PhD3, Boyd M.C. McCurdy, PhD3, Timothy van Beek3, Shashank Bhatia, PhD4 and Peter B. Greer,
Faculty of Science and IT, School of Mathematical and Physical Sciences, University of Newcastle,
NSW 2308, Australia 2
Faculty of Engineering and Built Environment, School of Electrical Engineering and Computer
M AN U
Science, University of Newcastle, NSW 2308, Australia 3
SC
1
RI PT
PhD1,4
Division of Medical Physics, CancerCare Manitoba, 675 McDermot Avenue, Winnipeg, Manitoba
R3E 0V9, Canada, Department of Physics and Astronomy, University of Manitoba, Winnipeg, Manitoba R3T 2N2, Canada, and Department of Radiology, University of Manitoba, Winnipeg, Manitoba R3T 2N2, Canada 4
TE D
Department of Radiation Oncology, Calvary Mater Newcastle Hospital, Locked Bag 7, Hunter region Mail Centre, Newcastle, NSW 2310, Australia
EP
Reprint requests to: Henry C. Woodruff, University of Newcastle, School of Mathematics and Physical Sciences, University Drive, Callaghan NSW 2308, AUSTRALIA; Tel: (+61) 432 911 085; E-mail: [email protected]
AC C
Conflict of interest: none.
ACKNOWLEDGEMENTS The authors gratefully acknowledge funding support from the Radiation Oncology Institute Grant# ROI2013-912, Cancer Institute New South Wales Hunter Translational Cancer Research Unit and the Calvary Mater Newcastle Radiation Oncology Research Fund.
ACCEPTED MANUSCRIPT Summary
AC C
EP
TE D
M AN U
SC
RI PT
This prospective clinical trial evaluated whether electronic portal imaging device images, in combination with a series of predicted portal images, can determine in real-time whether external beam radiation treatments are being delivered as planned. The data collected in this study will serve to determine action thresholds for future in-vivo real-time transit dosimetry software. Electronic portal transit image comparisons can simultaneously detect geometric and dosimetric treatment deviation stemming from patient and linear accelerator variations.
1
ACCEPTED MANUSCRIPT
ABSTRACT Purpose: Gantry-mounted megavoltage electronic portal imaging devices (EPIDs) have become ubiquitous on linear accelerators. WatchDog is a novel application of EPIDs, in which the image
RI PT
frames acquired during treatment are used to monitor treatment delivery in real time. We report on the preliminary use of WatchDog in a prospective study of cancer patients undergoing intensity
modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT), and identify the
SC
challenges of clinical adoption.
Methods and Materials: At the time of submission 28 cancer patients (head and neck, pelvis and
M AN U
prostate) undergoing fractionated external beam radiation therapy (24 IMRT, 4 VMAT) had at least one treatment fraction verified in real time (131 fractions or 881 fields). EPID images acquired continuously during treatment were synchronized and compared to model generated transit EPID images within a frame-time (~0.1 s). A χ-comparison was performed to cumulative frames to gauge overall delivery quality, and the resulting pass rates reported graphically during treatment delivery.
analysis.
TE D
Every frame acquired (between 500 and 1500 per fraction) was saved for post processing and
Results: The system reported mean ± 1 standard deviation real-time χ 91.1 ± 11.5% (83.6 ± 13.2%)
EP
for cumulative frame χ analysis with 4%, 4mm (3%, 3mm) criteria, global over the integrated image.
AC C
Conclusions: A real-time EPID based radiation delivery verification system for IMRT and VMAT has been demonstrated that aims to prevent major mistreatments in radiation therapy.
Keywords:
Real-time in-vivo dosimetry, EPID dosimetry, Intensity-modulated radiotherapy
2
ACCEPTED MANUSCRIPT INTRODUCTION External beam radiation therapy (EBRT) has evolved to become a complex delivery method for ionizing radiation. Intensity modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) lead the field of highly conformal delivery methods [1, 2].
RI PT
Due to the complex and non-intuitive fluences, often paired with large dose gradients, accuracy of delivery is usually verified by patient specific pre-treatment dosimetric
measurements [3-7]. However, a recent investigation at the Netherlands Cancer Institute
SC
revealed that nine of 17 serious errors identified in a cohort of 4337 patients would have been missed without in-vivo verification of radiation delivery [8]. Subsequently, in-vivo electronic
M AN U
portal imaging device (EPID) dosimetry for post-treatment analysis has been implemented by a few institutions [9-12]. With the increasing use of stereotactic body radiotherapy with very large doses delivered in 1-5 fractions, post-delivery analysis is not sufficient to detect and correct errors. We have developed and tested with phantom irradiations a real-time cine EPID
TE D
based verification system (WatchDog), enabling detection of gross treatment delivery errors prior to delivery of substantial radiation to the patient [13, 14]. A reference dataset of EPID images is predicted using parameters exported from the treatment planning system [15, 16],
EP
and compared to the stream of EPID images acquired during treatment. The comparison is
AC C
performed within a frame-time (~0.1 sec) and allows for both geometric (collimator positions, gantry angle, patient anatomy and positioning) and dosimetric (dose accuracy as a function of control point (CP) and gantry angle) verifications to be performed. Such a verification system provides an additional tool that can assist with prevention of major mistreatments in radiation therapy [10, 17-21].
In this paper we report on the first clinical demonstration of real-time EPID based dose delivery verification for patients undergoing IMRT and VMAT treatments. At the time of
3
ACCEPTED MANUSCRIPT submission, we have acquired EPID cine-images during treatment for 28 cancer patients (head and neck, pelvis, anal canal and prostate). The aims of this paper are to: 1) investigate accuracy of synchronization of measured and predicted image sets during patient deliveries; 2) determine prediction model accuracy for transit patient images using integrated image
RI PT
assessment; 3) determine initial results for real-time patient treatment verification with the system; 4) determine challenges to clinical adoption of the system.
SC
METHODS AND MATERIALS
M AN U
Overview and patient details
The study received ethical approval by the local human research ethics committee and the first phase was designed to be observational only, therefore no intervention was performed based on the monitoring information. The eligibility criteria were designed to ensure that
TE D
there would be a characteristic sample of treatment fields per patient. Twenty-eight patients with ages ranging from 51 – 78 years, with a median age of 67, had at least one fraction verified in real-time. All plans were hyper-fractionated and sites treated with IMRT were
EP
prostate (17 patients), head and neck (6 patients), and anal canal (1 patient) and four VMAT patients were treated in the pelvis area. The treatments were planned using Eclipse (Varian
AC C
Medical Systems, Palo Alto, CA) v11.
Linear accelerator, MLC, EPID, OBI and acquisition mode Two C-Series Trilogy linacs (Varian Medical Systems, Palo Alto, CA) equipped with Millennium 120-leaf MLCs were used to generate the 6 MV research (open air, phantom) and curative (patient) radiation fields. Megavoltage (MV) EPIDs (Varian PortalVision aS1000 flat panel detectors) were used to acquire the images, the detector square pixel having a side length of 0.0392 cm, yielding a total area of approximately 40 × 30 cm2. All data were
4
ACCEPTED MANUSCRIPT acquired in integrated acquisition mode by the clinical treatment software using the 4D Integrated Treatment Console (4DITC) PC. To capture image frames custom MATLAB/C# (MathWorks, Natick, MA, USA) code was developed in-house with a dual-base framegrabber card (Matrox Solios SOL 2M EV CLB), housed in a separate PC and connected via
RI PT
two camera-link cables to ports on the 4DITC and on-board imager (OBI) computers. The system captures every MV and kV frame, constituting a time-lapse series of individual MV frames and kV source angles acquired during dose delivery at frame rates of 7.455 fps and
SC
10.92 fps, respectively. Gantry angles were derived from kilovoltage (kV) source rotation information encoded in kV frames (the kV source was not active). Synchronization of the
M AN U
EPID image with the corresponding kV image containing the gantry angle information was achieved through time-stamps. All IMRT (VMAT) fields were clinical fields, with 400 (600) monitor units per minute (MU/min) nominal dose-rate respectively, and the EPID was positioned at 150 cm source-to-detector-distance (SDD). All EPID frames were dark-field
Prediction Model
TE D
and flood-field corrected by the 4DITC.
EP
For each treatment field a series of predicted EPID images were calculated at predetermined
AC C
meterset CP intervals (based on dose or gantry angle progression in IMRT or VMAT, respectively) using the method of Chytyk et al. [15, 16], providing a sequence of images for the entire beam delivery, as described in [13] and [14]. The model takes into account the energy response of the EPID by separating the energy-fluence into 15 energy bins, and incorporates support arm backscatter [22] and the attenuation and scatter caused by the patient and treatment couch in the beam, based on planning computed tomography (CT) data. The software automatically removes the CT couch from the CT volume and inserts the desired treatment couch. DICOM-RT files were used to input the treatment plan data into the
5
ACCEPTED MANUSCRIPT prediction model, and DICOM image files were used to import the CT volume for ray tracing. Artefacts in the CT volume introduced by patient implants also introduce artefacts in the predicted images, which were modeled at the rate of two (three) per CP for IMRT (VMAT) treatments to balance the need for a fine grid of predicted images with
RI PT
computational and storage limitations. The model was optimized for parameters such as intraleaf leakage (including tongue and groove effect), MLC leaf offsets, total backscatter and extra-focal contribution. Each predicted frame contains a distribution of grayscale values that
SC
are calibrated using a factor derived from the ratio of model prediction and measured EPID images at central axis for a 10x10 cm2 field at 150 cm SDD and 100 MU. The model was
M AN U
validated against measured EPID images for a wide variety of IMRT and VMAT test fields and clinical plans with and without phantoms present ([13], [14]). Image synchronization
The EPID acquires images at frame-rates set by the manufacturer and the image stream is not
TE D
synchronized to the linac’s dose delivery. Dynamically controlled dose-rate variations are varied “on the-fly” by the linac control systems and cannot be predicted. Therefore a method
EP
is needed to synchronize each measured frame to the corresponding predicted frame. For IMRT deliveries MLC leaf positions extracted from predicted and measured images are used
AC C
for synchronization, as described in [13]. The following changes were implemented in order to improve real-time processing: (i) the collimator angle was extracted from the treatment planning system (TPS) data instead of from the measured images; (ii) rather than rotating each image to 0° collimator angle to match the MLC leaf template, the template was rotated to match the collimator angle; (iii) the search space for predicted images available for synchronization was narrowed by moving boundaries (we selected 2 CPs behind and 10 CPs ahead of the currently synchronized CP). The accuracy of synchronization was assessed by comparing the synchronization index, which assigns each measured frame to the
6
ACCEPTED MANUSCRIPT corresponding predicted frame, with a reference step function (stepped due to the discrete number of CPs). This fiducial is derived after beam delivery from the ratio of the number of frames (“measured”/“predicted”).
RI PT
Due to the gantry-angle dependency of VMAT control systems, synchronization based on gantry-angles was chosen for VMAT deliveries, as described in [14]. The model uses TPS data to calculate EPID images at predetermined gantry angle increments. Measured EPID
M AN U
Real-time comparison and verification
SC
images are synchronized to the predicted images using gantry angle.
The system evaluates a cumulative image where each new measured image is added to the summed image and compared to the cumulative predicted image up to the current synchronization point, without normalization, yielding an evaluation which becomes less responsive to errors as a function of beam-on time but allows for dosimetric as well as
TE D
geometric evaluation of the treatment delivery [13]. Measured frames were binned to ½ resolution (512 x 384 pixels2) and 2-D comparisons were evaluated by using the χ method
EP
[24] to minimize the use of computing resources and perform all calculations within a frame time (~ 0.1s). The main aim of this study was to implement and gain initial experience with
AC C
the system and detect only major deviations from the treatment course, so a global criteria of 4%, 4 mm for pixels above a 10% of maximum signal threshold were used (dose criterion and threshold are calculated on the cumulative measured image), and a treatment field was flagged as failed if more than 4 consecutive frames fell below a χ pass-rate of 40%. These criteria were chosen from analysis of preliminary patient data, which showed large interfraction variation.
7
ACCEPTED MANUSCRIPT Clinical data acquisition The clinical workflow for a patient undergoing real-time verification differs little from a regular patient, with the presence of the extended EPID throughout treatment the only change apparent to the patient. The software has no ability to interfere directly with the treatment
RI PT
delivery. Prior to first fraction delivery treatment plan parameters and patient CT are exported from the TPS and the predicted frame set calculated. The predicted images are automatically encapsulated in a file containing all relevant patient and treatment parameters, which is then
SC
sent to the WatchDog PC at the linac. An integrated imaging template is attached to the
treatment plan to activate EPID image acquisition during delivery. Once the patient has been
M AN U
set-up for treatment, the radiation therapist (RT) selects the patient using the WatchDog graphical user interface (GUI) and presses “start” before the beam is delivered. The system automatically detects when the beam is on and automatically changes to next field when necessary. A simple “red / green light” flag and a progress bar are shown next to each field,
TE D
indicating delivery status. χ pass-rates are also reported graphically and in real-time. Figure 1 shows an example of the initial GUI implemented for patient treatment verification. All acquired clinical images and results were saved and system problems that prevented data
EP
acquisition were recorded and investigated. The real-time comparison produces a χ pass-rate
AC C
value for each measured image during the delivery of an IMRT field or VMAT arc. Here we present these results as the mean and standard deviation of these χ results. For each treatment site, the χ results are grouped for all patients and the mean and standard deviation quantified.
Figure 1: WatchDog graphical user interface for real-time treatment verification. The top left panel shows patient and treatment information, while the bottom left panel shows a progress
8
ACCEPTED MANUSCRIPT bar for each beam, the associated gantry angle, and a red/green status indicator. The plot in the upper right shows the real-time χ pass-rates and is updated with every new measured frame, with the measured EPID image displayed underneath.
RI PT
RESULTS Image synchronization
The accuracy of both IMRT and VMAT synchronization algorithms was previously found to
SC
be within 2 CPs (approx. 1-2% of total dose) when imaging through air or a phantom ([13, 14]). To assess the impact of variation in patient anatomy and/or positioning on
M AN U
synchronization accuracy 20 fields were measured during treatment with a patient present in the beam (10 each from H&N and prostate). These were subsequently synchronized to modeled frames with and without a patient CT scan included in the prediction model. The absence of the patient in the model was used to represent the effect of extreme anatomical
TE D
variation on the synchronization.
Figure 2 shows the χ pass-rates for the cumulative comparison for a prostate field, together
EP
with the associated synchronization indices, fiducial step function and their difference.
Figure 2: Prostate IMRT results for a single field for real-time verification (top),
AC C
synchronization indices (middle) and deviation from optimal synchronization in CPs (bottom) when comparing measured data with a patient to a model with the patient in the beam (left) and without (right).
The mean deviation (± standard deviation) of the synchronization index from the reference function was 0.04 ± 0.68 CPs including the patient compared to -0.07 ± 0.86 CPs without a patient in the model for the 20 patient deliveries. This shows that the synchronization
9
ACCEPTED MANUSCRIPT precision can be assumed to be independent of patient positioning or anatomical variation. The poor cumulative pass-rates for Figure 2(b) are due to the model not including the effect of the patient attenuation and scatter. VMAT deliveries were synchronized using measured
RI PT
and planned gantry angles which are not affected by imaging considerations [14].
Model and patient variability
The prediction model accuracy was validated against integrated patient transit EPID images
SC
acquired during treatments for a variety of IMRT (81 prostate, 43 head and neck and 14
pelvis) and VMAT (8 pelvis) fields for a total of 28 patients. Average ± 1 standard deviation
M AN U
integrated image χ pass-rates using 4%, 4mm (3%, 3mm) criteria, global over the integrated image, were found to be 96.7 ± 14.3% (92.7 ± 16.3)%, ranging from 61.4% to 100% (55.3 to 100%), with the largest divergence from the model found in prostate fields at 94.8 ± 15.2% (91.0 ± 15.9%) and the least divergence found in H&N fields at 98.1± 8.9% (95.6 ± 10.9%).
TE D
Delivery of two IMRT fields to one prostate patient in particular were responsible for all χ
EP
pass-rates < 80% and will be discussed separately as a case study.
Real-time error detection
AC C
The system results for real-time verification were analysed for the 28 patients. Total cumulative lag (i.e. lag at the end of each field) between delivery and reporting of χ pass-rates depends on the size of the delivered field as measured by the EPID (due to longer calculation times for the χ metric for larger fields), and was found to be between 0.3 s for smaller prostate fields and up to 0.9 s for the largest head and neck fields. Average cumulative χ passrates using 4%, 4mm (3%, 3mm) were 91.1% (83.6%) with a standard deviation of 11.5% (13.2%). Figure 3 summarises the data in a boxplot, sorted by treatment site.
10
ACCEPTED MANUSCRIPT Figure 3: Boxplot of real-time cumulative image χ passrates (4%, 4mm left, 3%, 3mm right, global) sorted by treatment site. The bottom and top of the box represent the first and third quartiles, and the separating line the median. The ends of the whiskers represent the
RI PT
minimum and maximum of all of the data for that category.
SC
Case study
A system which compares measured and predicted 2D transit EPID images will be sensitive
M AN U
to a range of dosimetric as well as geometric departures from the planned delivery, including internal anatomical changes. Error! Reference source not found. Mean real-time cumulative χ pass-rates (4%,4mm) for a prostate patient (7 field IMRT) over the course of 44 days (19 fractions) were 76.5% with a mean standard deviation of 6.3, ranging from (50.4-
TE D
94.8)%. No correlation was found between χ pass-rates and positioning metrics (couch parameters, set-up shifts, patient notes, set-up kV image analysis including gold seed and bone structure alignment) or patient weight. A single cone-beam CT was acquired in the last
EP
week of treatment that indicated large sections of the rectum filled with gas in contrast to the planning CT, which showed a rectum full of fecal matter. Moreover, increased organ motion
AC C
cannot be disregarded since prostate motion has been shown to increase in the presence of gas in the rectum (see, e.g. [25]), which could compound the departure from simulated conditions [26].
DISCUSSION A novel method for real-time verification of IMRT/VMAT treatment deliveries using EPID imaging was successfully implemented in a clinical setting in a cohort of 28 patients. Realtime EPID dosimetry complements routine quality assurance and presents a final safety net in
11
ACCEPTED MANUSCRIPT the EBRT process chain. As opposed to recent EPID-based in-vivo dosimetry solutions (discussed in detail in [9] and [10]), our system is designed to detect gross errors as they occur, before delivery of substantial dose to the patient, giving the therapists the option to
RI PT
interrupt a treatment and inquire into the causes of the discrepancy.
There are no precedents or recommendations to follow regarding real-time EPID transit
dosimetry. A global (over the integrated image) χ comparison criteria of 4%, 4 mm for pixels
SC
above a 10% threshold were used, and a treatment field would have been flagged as failed if more than 4 consecutive frames fell below a χ pass-rate of 40% for cumulative frame
M AN U
comparison. No errors, as per these parameters, were detected in the 131 fractions or 881 fields which were verified using the system, although low χ pass-rates prompted further enquiries into one patient (7 field prostate IMRT). A challenge to any image based analysis that distils the difference in dose distributions across two images down to one or two numbers
TE D
is how to distinguish between variations in the patient’s anatomy and changes in the delivered dose. Both cases imply that the dose was not delivered as planned, which is reported by our system, although false positives are possible in cases where large image differences (and
EP
hence low χ pass-rates) caused by, e.g. changes in the bony anatomy, may lead to small
AC C
differences in the dose delivered to the target volume. Optimal χ criteria will be the objective of further studies based on larger patient cohorts. These will be designed to detect clinically relevant errors and not small errors owing to transient anatomical changes [26-28]. The system is not able to verify fields where collisions between EPID and the couch or the patient are possible, such as non-coplanar beams. Patient with implants, which create large artefacts in the planning CT volume and also lead to large artefacts in the predicted EPID images, were excluded from this study
12
ACCEPTED MANUSCRIPT There are several issues for clinical adoption that have been identified in these preliminary patients. The system uses a research computer that interfaces with the clinical imaging systems via a third party frame-grabber card, and there are challenges that are being overcome to make this interface robust. The forward prediction model takes up to 12 hours to
RI PT
produce a stream of predicted EPID images, but these calculations can be performed in
batches and unsupervised, and the software supports dedicated graphical processing units, which can halve prediction times. There is currently no method to change the order of the
SC
fields as they are delivered, a feature that the RTs involved in this study have asked to be introduced in future versions. Image acquisition can also be missed, as the system must be
M AN U
separately activated before treatment although the frequency of this decreased throughout the study. Cone-beam CT (CBCT) acquisition stops the stream of kV dark frames to the framegrabber, presenting a problem specific to VMAT deliveries that require constant update of the gantry angles. The current solution is to add a single kV image setup-field (e.g. anterior-
CONCLUSION
TE D
posterior) in addition to the CBCT, which must be moded up just prior to treatment.
EP
We have presented the successful clinical implementation of WatchDog, an independent system to verify EBRT treatment in real-time using EPIDs. The frame-stream is synchronized
AC C
to the predicted images and evaluated using the χ comparison. Average real-time cumulative χ global pass-rates using 4%, 4mm (3%, 3mm) over the integrated image were 91.1% (83.6%) with a standard deviation of 11.5% (13.2%). A clinical case study involving a patient with gas in the rectum illustrated the system’s ability to indicate deviations from the original plan that would have otherwise remained undetected.
13
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
REFERENCES 1. Yu, C.X., Intensity-modulated arc therapy with dynamic multileaf collimation: an alternative to tomotherapy. Physics in Medicine and Biology, 1995. 40(9): p. 1435. 2. Otto, K., Volumetric modulated arc therapy: IMRT in a single gantry arc. Medical Physics, 2008. 35(1): p. 310-317. 3. Both, S., et al., A study to establish reasonable action limits for patient specific IMRT QA. 2007. Vol. 8. 2007. 4. van Elmpt, W., et al., The next step in patient-specific QA: 3D dose verification of conformal and intensity-modulated RT based on EPID dosimetry and Monte Carlo dose calculations. Radiotherapy and Oncology, 2008. 86(1): p. 86-92. 5. Ling, C.C., et al., Commissioning and Quality Assurance of RapidArc Radiotherapy Delivery System. International Journal of Radiation Oncology*Biology*Physics, 2008. 72(2): p. 575-581. 6. Van Esch, A., et al., Implementing RapidArc into clinical routine: A comprehensive program from machine QA to TPS validation and patient QA. Medical Physics, 2011. 38(9): p. 5146-5166. 7. O'Daniel, J., et al., Volumetric-modulated arc therapy: effective and efficient end-toend patient-specific quality assurance. Int J Radiat Oncol Biol Phys, 2012. 82(5): p. 15671574. 8. Mans, A., et al., Catching errors with in vivo EPID dosimetry. Medical Physics, 2010. 37(6): p. 2638-44. 9. van Elmpt, W., et al., A literature review of electronic portal imaging for radiotherapy dosimetry. Radiotherapy and Oncology, 2008. 88(3): p. 289-309. 10. Mijnheer, B., et al., In vivo dosimetry in external beam radiotherapy. Medical Physics, 2013. 40(7): p. 070903. 11. Berry, S.L., et al., Implementation of EPID transit dosimetry based on a through-air dosimetry algorithm. Medical Physics, 2012. 39(1): p. 87-98. 12. Berry, S.L., et al., Initial clinical experience performing patient treatment verification with an electronic portal imaging device transit dosimeter. Int J Radiat Oncol Biol Phys, 2014. 88(1): p. 204-9. 13. XXXXX 14. XXXXX 15. Chytyk, K. and B.M.C. McCurdy, Comprehensive fluence model for absolute portal dose image prediction. Medical Physics, 2009. 36(4): p. 1389-1398. 16. Chytyk-Praznik, K., et al., Model-based prediction of portal dose images during patient treatment. Medical Physics, 2013. 40(3): p. 031713. 17. Mayles, W.P.M., The Glasgow Incident — a Physicist's Reflections. Clinical oncology (Royal College of Radiologists (Great Britain)), 2007. 19(1): p. 4-7. 18. Williams, M.V., Radiotherapy Near Misses, Incidents and Errors: Radiotherapy Incident at Glasgow. Clinical oncology (Royal College of Radiologists (Great Britain)), 2007. 19(1): p. 1-3. 19. Derreumaux, S., et al., Lessons from recent accidents in radiation therapy in France. Radiat Prot Dosimetry, 2008. 131(1): p. 130-5. 20. Ortiz Lopez, P., et al., ICRP publication 112. A report of preventing accidental exposures from new external beam radiation therapy technologies. Ann ICRP, 2009. 39(4): p. 1-86. 21. Bogdanich, W. Radiation offers new cures, and ways to do harm. The New York Times, 2010. 22. Rowshanfarzad, P., et al., Measurement and modeling of the effect of support arm backscatter on dosimetry with a Varian EPID. Medical Physics, 2010. 37(5): p. 2269-2278.
14
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
23. XXXXX 24. Bakai, A., M. Alber, and F. Nüsslin, A revision of the γ-evaluation concept for the comparison of dose distributions. Physics in Medicine and Biology, 2003. 48(21): p. 3543. 25. Ogino, I., et al., Reduction of Prostate Motion by Removal of Gas in Rectum During Radiotherapy. International Journal of Radiation Oncology*Biology*Physics, 2008. 72(2): p. 456-466. 26. Kroonwijk, M., et al., In vivo dosimetry for prostate cancer patients using an electronic portal imaging device (EPID); demonstration of internal organ motion. Radiotherapy and Oncology, 1998. 49(2): p. 125-132. 27. McDermott, L.N., et al., Replacing Pretreatment Verification With In Vivo EPID Dosimetry for Prostate IMRT. International Journal of Radiation Oncology*Biology*Physics, 2007. 67(5): p. 1568-1577. 28. Persoon, L.C., et al., Interfractional trend analysis of dose differences based on 2D transit portal dosimetry. Phys Med Biol, 2012. 57(20): p. 6445-58.
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT