EPIDose : An Overview of EPIDose and the EPIDose Process and Algorithm SUN NUCLEAR corporation Your Most Valuable QA and Dosimetry Tools
introduction Pre-treatment dose QA is an important process required for every patient receiving complex beam designs, such as intensity modulated radiation therapy (IMRT) and volume modulated arc therapy (VMAT). Dose QA helps to ensure that: 1) The treatment planning system (TPS) has calculated each beam accurately, and 2) The delivery system is capable of delivering each beam accurately. The complex and unique nature of IMRT beams requires that each beam s dose be verified thoroughly over its full cross-sectional area. To adequately verify a beam, a dosimetry system must be both high resolution and high density. High resolution allows the complex, high-gradient dose values to be measured without volume-averaging effects that blur the dose profiles; high resolution is achieved with detectors that have a small (< 1 mm x 1 mm) active area, as seen by an incident beamlet. High density ensures that enough data points are audited to catch even small regions of errors, as small regions of high error may be clinically relevant even though they may have only a minor effect on the conventional passing rates often used in QA. 2D diode arrays were introduced approximately 10 years ago, and they offered an efficient per-beam dosimetry solution (1, 2) that had high resolution and adequate density. More recently, the mega-voltage (MV) Electronic Portal Image Device (EPID) has received attention in the realm of perbeam dose QA. The MV EPID was originally designed for image-guidance purposes, i.e. to generate quick anatomic images at the treatment beam geometry to compare with digitally-reconstructed radiographs (DRR) or simulation films to verify patient alignment. (3) However, the high data density afforded by the EPID, along with the convenience of having the detector panel attached to the treatment machine opposite the beam, incited physicists to investigate the EPID s potential usefulness for per-beam dose QA. (4-10) 2 In this paper, we discuss the basics of a commercial system (EPIDose, Sun Nuclear Corporation, Melbourne, FL) that generates absolute dose plane estimates, using raw EPID image input, for usage in per-beam dose QA. Studies on the performance of EPIDose have been published previously. (10,11) The purpose of this paper is to summarize in more detail how EPIDose works. Benefits of EPIDs for Dose-Related Applications Overview In discussing the specifics of the EPIDose algorithm, we must first itemize the limitations of the EPID, but even before that, it is useful to review why there is interest in using EPID for dose QA purposes. As mentioned previously, the two major strengths of the EPID are: 1) data density, and 2) convenience of setup. Data Density Typical commercial EPID panels are about 30 cm x 40 cm large, and can be positioned in the beam path at a range of distances. The panel itself generates images with pixels < 1 mm in size, thus giving high resolution images without volume averaging. This is a resolution similar to a diode array. (1) However, unlike arrays comprised of ion chambers or diodes, the EPID has a continuous pattern of pixels with no gaps in between, thus generating an optimal data density. The EPID data density thus captures the beam profile completely, which is very useful when probing steep and complex dose gradients that are a characteristic of IMRT. Figure 1 gives an example of the type of analysis allowed by EPIDose vs. the same analysis done with a lowerdensity array. Convenience of Setup The EPID itself is attached to the linear accelerator as a retractable arm. Thus, the setup of the EPID opposite the treatment beam is automated, requiring no human lifting and alignment of a phantom either on the couch or in a mounting fixture. This convenience can save a little time for the clinical physicist in his/her QA routine.
[A] [B] Figure 1 EPID images have high data density (continuous) and high resolution (sub-mm input pixels), which allow for the careful analysis of complex dose gradients. As seen in this figure, the low density diode array (panel A) shows a pattern of errors with high resolution, but the EPIDose analysis (panel B) exhibits high resolution and high density, making the high gradient errors in small regions to be easily seen. Notice the sensitivity even to tongue-and-groove effects (red and blue horizontal stripes of error). 3
the challenges of using EPIDs for dose-related applications Despite the potential benefits of the EPID for doserelated applications, it is imperative to acknowledge that the EPID was not designed as a dosimeter. It does not measure dose, and in fact there are many characteristics of EPID response that deviate from dose response in water or near water-equivalent media. These characteristics of EPID response cause difficulties and challenges for those wishing to use EPID for dose-related studies, and these challenges must be overcome by any system that claims to allow EPID-based dosimetry. Some (not all) of the challenges of EPID-based dosimetry are summarized here: EPID pixel profiles are not equivalent to dose profiles (see Figure 2). EPID response to variable energy spectra, including variations caused by head and MLC-scattering, is not equivalent to dose response. EPID response is often asymmetric due to the panel mechanics and/or software (see figure 3). EPID response near the edges of the active area can be inaccurate and unreliable (see figure 4). EPID image output is highly dependent on calibration and software adjustments built into the system, adjustments which if not done, or done incorrectly, can result in bad data (see figure 5). Constancy of EPID response can vary with time and will change with manipulations in the EPID hardware or software (such as after servicing is performed, or upgrades, etc.), and this has an impact on dose calibration. EPID alignment (i.e. the location on the panel where the CAX intersects) can shift with gantry angle. (12) EPID response does not reflect lateral scatter of dose in-tissue (or in-phantom). Raw EPID Dose at depth Figure 2 EPID image pixel profiles (even those rendered in calibrated units for portal dosimetry) do not match dose profiles taken at depth in water-equivalent (or similar) media. This is true for any chosen depth in media. This reinforces that EPID images by themselves (i.e. those not processed by an algorithm such as EPIDose) cannot be used to audit absolute dose calculation by a TPS or dose delivery by a linac. Data shown here are from a Varian EPID. 4
Figure 4 This example shows a large IMRT segment which overlaps one of the extreme Y borders of the EPID (panel A). The resulting pixel profile clearly shows a large over-response at that border (panel B), a response which is not seen in the dose profile of this same segment. Data shown here are from an Elekta EPID. [B] [A] Figure 3 EPIDs often exhibit asymmetric response even to symmetric open field delivery. This example shows a simple 20 cm x 20 cm square field where the EPID image is symmetric in X but asymmetric in Y. Degrees of asymmetry errors can be mild (as shown here) to quite large. In addition, the degree of asymmetry (the slope) often changes with field size for open fields. Data shown here are from an Siemens EPID. Figure 5 Unless EPID hardware and software systems are calibrated properly and adjusted for background and flood fields, the resulting pixel response can be very noisy as seen here. This noise is not a result of true noise in dose, and thus is problematic if a user wishes to use EPIDs for any dosimetry-related applications. 5
the EPIDose solution Overview Dose QA for IMRT/VMAT plans requires auditing of both: A) the TPS dose calculation and B) the delivery system. In order to audit the TPS dose calculation, it is obvious that a calculated dose must be generated so that any error or imperfection in the TPS would manifest itself (errors due to patient heterogeneities not included). Other EPID-based QA systems compare a predicted image (derived from the TPS beam fluence and an image generation algorithm) with the measured image, and thus the TPS dose engine is not audited. EPIDose, on the other hand, works exactly like dose QA using a dose measurement array such as MapCHECK, i.e. a measured absolute dose plane is compared to a calculated absolute dose plane. It is the job of EPIDose, then, to accurately simulate a measured absolute dose plane to mimic a conventional measurement (with MapCHECK, for instance) at a specific dose distance and equivalent depth in a flat phantom. Now, we have covered many of the reasons why EPID images cannot simply be calibrated or converted to true dose planes by scaling. It is therefore evident that an algorithm must be employed that corrects for all the nuances and physical challenges of the EPID that make it differ from true dosimetry. With EPIDose, a physics model is optimized that is used in conjunction with the EPIDose algorithm to generate absolute dose planes to mimic a flat dosimetric array. The EPID itself may be at any distance from the source (i.e. Figure 6 This figure shows the detection of a TPS beam modeling error for a 15 MV beam. The top panel (A) is analyzed by the MapCHECK diode array. The bottom panel (B) is analyzed by EPIDose. The high data density of EPIDose makes the distribution of errors easier to see, enabling the physicist to diagnose and fix the beam modeling error. NOTE: The specific beam modeling here resulted from the usage of ion chambers to scan measured dose profiles in the Eclipse TPS, resulting in pencil beam gradients that were adversely affected by the volume-averaged penumbra. This phenomenon has been described in literature. (13) 100cm, 140cm, 160cm, etc.) and no build-up needs to be placed atop the EPID panel. The dose plane generation can be for a user-specified preference (e.g. 100cm dose plane distanced, 10cm equivalent depth). An EPIDose physics model will vary with EPID distance, QA dose plane geometry, MLC model, and beam energy; however, most users will prefer to use constant geometry in their process anyway, meaning that, in practice, there can be one EPIDose model per linac, per energy. Analysis with EPIDose enables the physicist to take advantage of the superior data density of the EPID pixel array. This can be very valuable when one is optimizing a TPS beam model or doing delivery QA on a delivery system. Figure 6 gives an example of how data density can help a physicist diagnose errors and thus be directed in how to mitigate the errors. The work-flow of EPIDose is detailed in Figures 7-8. There is a slightly different process for the Varian and Siemens systems (Figure 7) as there is for the Elekta system (Figure 8). Elekta IMRT EPID images are captured oneper-segment, whereas the Varian and Siemens images are capture one-per-beam. This affords certain advantages to EPIDose with Elekta (improvements in output factor and transmission region correction, for example), though the number of input image files is clearly more with Elekta (one per segment). There is a separate EPIDose module to automate the process using Elekta EPID log files. [A] [B] 6
Figure 7 Work- and data-flow of EPIDose when used with Varian or Siemens EPID systems. Figure 8 Work- and data-flow of EPIDose when used with the Elekta EPID system. 7
algorithm This section describes the basics of the EPIDose algorithm. Descriptions are often more clear when images and/or schematics can be used, so we will rely on Figures 9-13 heavily. Prior publications (10) have outlined the algorithm and provided useful figures, but here we offer more detail. The EPIDose algorithm has distinct stages, and these have been schematically detailed in Figure 9. In summary, the stages are: A) geometry corrections, B) per-segment output corrections (both inside the exposed areas and under MLC leaves, corrected separately), C) dose profile convolution/ redistribution, and D) absolute dose calibration over a wide field. Step A is a geometric scaling, allowing the raw EPID image (acquired at some distance) to be scaled to the distance of the desired dose plane. The user chooses the parameters that suit their needs, and for any EPIDose model the parameters are fixed. Some users like to maximize the area coverage of the EPID by moving it up to the isocenter distance (100cm). Others, and commonly Elekta users, leave the EPID at a larger distance for convenience (say, 160 cm), which decreases the effective field size that can be captured by the panel (30cm panel dimension at 160cm projects to 18.75cm at isocenter). Step B (per segment output correction) is built around using 2D output factor correction factor (OFCF) maps (Figure 10). This process is fairly straight-forward for Elekta EPIDs, as these EPID images are acquired and saved per-segment, offering direct pixel manipulation to each component image prior to generating a composite dose plane. However, for Varian and Siemens EPID, the EPID image itself is already composited, and therefore a 2D correction map must be built using a priori knowledge of the IMRT segment sizes, shapes, and relative weights. This information comes from the RT Plan, where each beam is described as a series of control points of machine parameters. Output corrections are different for exposed regions vs. transmission regions, and the OFCF generator must handle these conditions over all segments and for both SMLC and DMLC delivery. Figure 11 illustrates the basics of this process, using a fictitious IMRT beam comprised of three simple segments. Though clinical IMRT beams have typically between 5 and 200 segments of complex MLC shapes, the mechanics illustrated in Figure 11 still apply. Steps C and D are described by Figures 12 and 13, respectively. A caveat to Step D (calibration) is that EPIDose allows the user to truncate (i.e. ignore) certain border pixels near the edge of the EPID if the user prefers, and this is to mitigate the strange and inaccurate pixel values that can occur in these regions (Figure 4). Figure 9 EPIDose is an algorithm that renders estimated absolute dose planes from input EPID images. The algorithm has several serial stages: A) geometric manipulation, B) correction for output factor vs. equivalent field sizes per segment, C) dose redistribution to give equivalent profile as dose-at-depth, and D) conversion to absolute dose using a 2D calibration map. The resulting EPIDose planes can be used to: 1) commission/optimize the EPIDose model (by comparing to absolute dose planes of a diode array or other method), or 2) perform IMRT QA by comparing to TPS dose planes, calculated at the simulated QA depth in phantom. 8
Figure 10 An IMRT field is a series of several to many individual segments, delivered statically (step-and-shoot, or SMLC) or dynamically (DMLC). Each segment can require a different correction to adjust the EPID response to be dose-equivalent at depth. Therefore, the IMRT segment information (MU weights and MLC shapes) must be processed to generate the output-factor correction factor (OFCF) map, which is applied to the input data. Figure 11 A simplified, illustrative example of calculating an OFCF map for a fictitious three-segment field. The dark gray represents the open portion of the MLC while the light gray is the transmission region. Figure 13 The processed data are rendered to absolute dose planes by a wide field calibration routine. Figure 12 A dose redistribution kernel is applied to the input data to render the dose profiles to be doseequivalent at depth in phantom. For 6MV beams at simulated QA depths that are shallow, the kernel is usually very narrow; but the width increases with increasing simulated depth and more so for higher beam energies. The example shown here is exaggerated for emphasis. 9
physics modeling Physics modeling of EPIDose is summarized in the flow chart in Figure 14. Physics modeling is broken into two phases: Phase I - in which data integrity is verified and modeling of simple, open fields is established, and Phase II - which is an iterative process of finding the optimal parameters to generate EPIDose results that fit an independent measurement system (MapCHECK or MapCHECK2) for a variety of IMRT fields/plans. Often, Phase II optimization will cause a model to deviate from its optimal open field model, and it is not uncommon to store a model for open field analysis and a separate model for IMRT analysis. Figures 15-16 show example results of well-commissioned EPIDose beam models. In these figures, EPIDose results are shown using absolute dose analysis vs. MapCHECK dose planes. Of course, when EPIDose is commissioned, then per-patient dose QA is performed by comparing EPIDose vs. TPS calculated dose planes. Figure 14 EPIDose modeling requires two phases. The Phase I goal is to model simple open fields and verify data validity (EPID linearity of response, smoothness and constancy of data, etc.). Once Phase I is complete and objectives met, Phase II is undertaken and consists of optimizing the EPIDose model using field-by-field analysis of IMRT beams for a range of treatment plans. NOTE: A model optimized for IMRT may deviate from a model optimized for simple open fields, and often two models will be stored. 10
Figure 15 A properly modeled EPIDose system will generate dose profiles that match the MapCHECK dose profiles very well. The MapCHECK dose is measured at the QA depth, while the EPID images are acquired with no build-up at the EPID distance, then rendered to dose-equivalent planes at the simulated QA depth and distance. Figure 16 This example shows EPIDose absolute dose plane results that match MapCHECK 2 absolute dose (5cm equivalent depth, 100cm dose plane distance) with a gamma passing rate of 100.0% using 2%/2mm gamma criteria (20% lower dose threshold cutoff, global % normalization method). 11
references 1 A 2-D diode array and analysis software for verification of intensity modulated radiation therapy delivery, Jursinic, P.A. & Nelms, B.E.Medical Physics 30, 870 (2003). 2 Evaluation of a 2D diode array for IMRT quality assurance, Letourneau, D., Gulam, M., Yan, D., Oldham, M. & Wong, J.W. Radiother. Oncol. 70, 199 206 (2004). 3 Electronic portal imaging devices: a review and historical perspective of contemporary technologies and research, Antonuk, L.E.Phys. Med. Biol. 47, R31-R65 (2002). 4 Dosimetric properties of an amorphous silicon electronic portal imaging device for verification of dynamic intensity modulated radiation therapy, Greer, P.B. & Popescu, C.C. Medical Physics 30, 1618-1627 (2003). 5 Dosimetric IMRT verification with a flat-panel EPID, Warkentin, B., Steciw, S., Rathee, S. & Fallone, B.G. Medical Physics 30, 3143-3155 (2003). 6 The use of an asi-based EPID for routine absolute dosimetric pretreatment verification of dynamic IMRT fields, Van Esch, A., Depuydt, T. & Huyskens, D.P. Radiother. Oncol. 71, 223 234 (2004). 7 The impact of MLC transmitted radiation on EPID dosimetry for dynamic MLC beams, Vial, P., Greer, P.B., Hunt, P., Oliver, L. & Baldock, C. Medical Physics 35, 1267-1277 (2008). 8 Portal dose image prediction for in vivo treatment verification completely based on EPID measurements, Van Zijtveld, M., Dirkx, M., Breuers, M., de Boer, H. & Heijmen, B. Medical Physics 36, 946-952 (2009). 9 A simple approach to using an amorphous silicon EPID to verify IMRT planar dose maps, Lee, C., Menk, F., Cadman, P. & Greer, P.B.Medical Physics 36, 984-992 (2009). 10 Evaluation of a fast method of EPID-based dosimetry for intensitymodulated radiation therapy, Nelms, B. E., Rasmussen, K. H., & Tomé, W. A. Journal of Applied Clinical Medical Physics 11(2), 140-157 (2010). 11 EPID dosimetry for pretreatment quality assurance with two commercial systems, from IMRT to VMAT, Bailey, D. W. et al. (Currently under review) 12 Using an EPID for patient-specific VMAT quality assurance, Bakhtiari M., Kumaraswamy L., Bailey D. W., de Boer S., Malhotra H. K., & Podgorsak M. B. Medical Physics, 38(3), 1366-1373. 13 The use of film dosimetry of the penumbra region to improve the accuracy of intensity modulated radiotherapy, Arnfield, M. R., Otto, K., Aroumougame, V. R., & Alkins, R. D. Medical Physics, 32(1), 12-18 (2005). SUN NUCLEAR corporation Your Most Valuable QA and Dosimetry Tools Corporate Headquarters 425A Pineda Court Melbourne, Florida 32940 tel: + 1 321 259-6862 web: www.sunnuclear.com 1214W06012011