| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Full Paper |
1 Joint Department of Physics, The Institute of Cancer Research and the Royal Marsden NHS Trust, Downs Road, Sutton, Surrey SM2 5PT and 2 National Physical Laboratory, Centre for Ionising Radiation Metrology, Dosimetry Group, Kaye Building, Queens Road, Teddington TW11 0LW, UK.
 |
Abstract |
|---|
 Top Abstract IntroductionMethods and materialsResultsDiscussionReferences |
|---|
Key Words: 1(i)(j)(a)
|
Introduction |
|---|
|
TopAbstractIntroductionMethods and materialsResultsDiscussionReferences |
|---|
Since the patterns of treatment differ in Europe from the US, and since there are a large number of centres in the UK using Elekta linear accelerators, the experience gained by the use of Pinnacle3 in the US is not necessarily applicable in the UK [1, 2]. Some degree of familiarization and experience is therefore required. Accordingly, this paper presents details of the commissioning and quality assurance of Pinnacle3 at the Royal Marsden NHS Trust. It is not intended that this should be a guideline or recommendation for the use of Pinnacle3, but that other users or potential users may benefit from the approach taken for the commissioning process.
Several papers are of relevance to the present work. The most comprehensive guideline is the report by the Radiation Therapy Committee Task Group 53 (TG53) of the American Association of Physicists in Medicine on quality assurance for clinical radiotherapy treatment planning [3]. This is a complete description of the test procedures that should be carried out when commissioning a new treatment planning system. The paper by Van Dyk et al [4] is somewhat less involved and more practically realizable. Venselaar et al [5] have also provided a comprehensive list of tolerances for acceptance of treatment planning system accuracy, and applied these criteria to systematic tests of a variety of treatment planning systems, including Pinnacle3 [6]. Several other authors have reported on aspects of the commissioning process [7–10]. In particular, Starkschall et al [11] have described the beam modelling procedure used in conjunction with Pinnacle3, but have not discussed the complete set of tests required fully to commission the treatment planning system. Thus, the present paper provides a broader overview of the commissioning process for Pinnacle3, as carried out at this centre.
|
Methods and materials |
|---|
|
TopAbstractIntroductionMethods and materialsResultsDiscussionReferences |
|---|
For a separate small-field beam model (see below), dose profiles were measured using the above-mentioned diode and/or film for field sizes down to 2 cm. Output factors were measured using a 0.13 cm3 ionization chamber (Wellhöfer, Schwarzenbruck, Germany) down to a 1 cm field.
CT-to-density table
An RMI 465 density phantom (Radiation Measurements Inc., Middleton, WI) was used to generate CT-to-density tables at the 120 kV and 140 kV CT voltages commonly used at this centre [10]. These CT-to-density tables were used for all subsequent testing. An in-house CT quality assurance phantom [12] was also scanned. This phantom was a cylindrical water-filled phantom, 25 cm in diameter and 16 cm in length, containing 6 eccentric cylinders, each of 5 cm diameter, filled with various materials. These materials included water, air, lung-equivalent material (0.28 g cm-3) and two densities of bone-equivalent material (1.35 g cm-3 and 1.85 g cm-3). A central cavity allowed a 0.6 cm3 ionization chamber to be inserted into the phantom along its axis. This phantom was used for periodical verification of the CT-to-density tables and to perform dosimetric tests (see below).
Pinnacle3 beam model
The measured data were imported into Pinnacle3 using the import facilities provided. The Pinnacle3 approach was then to use the measured data to generate a model of the accelerator head, from which dose in any arbitrary situation could be calculated. The principal elements of the beam model were as follows [11, 13]:
Use of Monte Carlo generated accelerator data
Most of the above parameters represented physical features of the linear accelerator head. It was therefore possible to use the data generated from a Monte Carlo model of the accelerator as a starting point for these parameters. Hence, a Monte Carlo model was derived from measured dose distributions in water using the approach described by De Vlamynck et al [14]. The energy distribution and the lateral width of the primary electron beam hitting the target were obtained by matching calculated and measured dose distributions in water. The Pinnacle3 parameters were extracted from the Monte Carlo simulation data. The resulting dose distributions in water were compared with measurements [15], and the Pinnacle3 parameters were, where necessary, manually adjusted to provide the best possible fit of the dose distributions to the measurements.
Automodelling
The option within Pinnacle3 to generate beam parameters automatically according to measured data was also used. For each combination of linear accelerator and beam energy, separate models were produced for open and wedged fields. The Pinnacle3 sequence D_TuneAllInSections was first run to produce an approximate model [11, 13]. This process tuned the energy spectrum of the incident photon beam and the electron contamination parameters, while adjusting the entire model. The dimensions of the focal spot were also adjusted by this process. The process FineTuneECAndSpect was then run to improve the energy spectrum and electron contamination characteristics simultaneously. The process FineTuneCrossBeam was also run, to improve the cross-beam profiles for a range of field sizes. For the wedge model, processes FineTuneAllForWedge and FineTuneModifierScatter were both used to provide additional accuracy. Output factors were then computed.
Small-field beam model
The treatment planning system was required for stereotactically guided conformal radiotherapy of benign brain tumours, where multiple static beams were to be delivered with small non-coplanar fields. The beam data used by the planning system to plan these precise treatments was required to be highly representative of the treatment machine beams for the range of field sizes used. At this centre, an in-house conformal blocking system is used, which improves collision-free beam access around the patient compared with the standard accessory tray. Given the differences between this system and the tray-mounted blocks, in addition to the need to extend the field size range down to 2 cm and the greater beam modelling accuracy required for high-precision brain treatments, a separate small-field beam model was desirable. Therefore, small-field 6 MV treatment machine data incorporating two beam models, manually optimized for the field size range 2 cm to 12 cm, were created on Pinnacle3. A 0.15 cm grid resolution was used for the modelling. The two models used the same Monte Carlo generated spectrum as above. However, a different combination of modelling parameters was arrived at, which also produced reasonable agreement with measurements.
Dose profiles and blocked field output factors for various regular and irregular conformal blocks were compared with measurements. A circular block aperture of diameter 3.7 cm and a square block aperture of side 3.7 cm were both tested with secondary jaw sizes of 4.5 cm x 4.5 cm. Also tested were two blocks with irregularly shaped apertures previously used for a patient's treatment. These had approximately 40% of the secondary beam area shielded, the field dimensions ranging from 5.5 cm to 7.5 cm, one beam being symmetrical and the other asymmetrical. In order to remove the block production and light/radiation field discrepancies, each block shape was digitized into Pinnacle3 from the 50% isodose contour obtained using film. Accordance with the TG53 example criteria [3] was taken as a minimum requirement for validation of the model.
Assessment of heterogeneity correction
Several tests were carried out to verify the heterogeneity correction in a similar manner to that described by Kappas and Rosenwald [16] and Rice et al [17]. To test the fundamental performance of the heterogeneity correction, blocks of the same lung- and bone-equivalent material as used in the in-house CT phantom (see above) were positioned between slabs of "solid water" (Radiation Measurements Inc., Middleton, WI). A thickness of 5 cm of solid water was used for build-up in all cases. The thickness of the lung-equivalent material was 10.0 cm, whilst that of the 1.35 g cm-3 bone was 8.7 cm and that of the 1.85 g cm-3 bone was 8.6 cm. Approximately 15 cm of solid water was positioned beyond the inhomogeneity.
By dividing the solid water and inhomogeneous material into two widths, it was possible to sandwich an extended dose range verification film (EDR2, Eastman Kodak, Rochester, NY) through the cross-section of the material. This was then irradiated with 5 cm x 5 cm fields of energy 6 MV and 10 MV to provide a depth dose curve through the inhomogeneous material. The gantry was angled at 5° so as to minimize the perturbation caused by the presence of the film through the inhomogeneity. The films were calibrated using a similar geometry, but with solid water only. The depth dose curves were normalized to the dose at 3 cm in depth, so as to avoid any instability at the depth of maximum dose, but while sufficiently shallow to avoid being affected by the inhomogeneities. The depth doses obtained in the presence of inhomogeneities were divided by the corresponding depth dose in water, to provide a correction factor representing the influence of the inhomogeneity, following the benchmark measurements of Rice et al [17].
Measurements with a 0.6 cm3 ionization chamber were also made at 2 cm, 5 cm and 10 cm beyond the inhomogeneity. These measurements were made for gantry angles of both 0° and 5°, in the absence of film. Corresponding measurements were also made in homogeneous solid water, so that the ratio of the measurements with and without heterogeneity could be used as the correction factor.
Within Pinnacle3, depth dose curves were created using the collapsed cone convolution algorithm on a 2 mm x 2 mm x 2 mm grid. Correction factor curves were created by taking the ratio of heterogeneous and homogeneous curves.
To provide a more clinically applicable assessment of the inhomogeneity correction within Pinnacle3, a variety of six-field treatment plans were created and delivered to the in-house CT quality assurance phantom [12], and the resulting isocentric dose measured for each field using the ionization chamber. The planned and delivered doses were then compared. For this test, 5 cm x 5 cm fields were used so as to correspond with the 5 cm diameter inserts in the phantom.
Assessment of dosimetry for oblique, asymmetric and irregular field shapes
Various water phantoms were used to check the absolute accuracy of the treatment planning system in calculating oblique, asymmetrical and irregular fields. These fields were planned and delivered to the phantoms, and an ionization chamber was used to check the absolute dose. In addition, verification film was used to determine the correspondence of the isodoses predicted by the treatment planning system for MLC and block penumbra, with the corresponding measurements.
Quality assurance of dose–volume histograms
For the assessment of Pinnacle3's dose–volume histograms (DVHs), three methods were used: (1) calculation for the case of a precisely known dose distribution; (2) variation of the grid size; and (3) comparison with an independent algorithm. For the first of these tests, a cubic phantom of side 30 cm was defined, containing a cubic volume of side 6 cm at the centre. A dose distribution consisting of eight cubic octants, each with a dose of between 10 Gy and 80 Gy, in discrete intervals of 10 Gy, was created externally from the planning system using custom software. The dose distribution was then imported into the planning system. The DVH was computed on a grid with 4 mm x 4 mm x 4 mm spacing, which was a typical clinical setting. Since the volume of interest was positioned centrally, the expected result was that one eighth of the structure would receive 10 Gy, one eighth would receive 20 Gy, and so on. The structure was then shifted in various directions by either 2 mm or 4 mm, to observe the influence on the DVH of the position of the structure relative to the dose grid. Although this was an artificial test, it enabled a thorough examination of the DVH algorithm, providing greater precision than was possible by simple interaction with the treatment planning system. This type of approach was taken by Jacky and White [18] for testing a complete radiotherapy treatment planning system.
The second test was to contour three concentric spherical volumes of interest, with diameters of 6 cm, 8 cm and 10 cm, respectively. Four treatment fields, each of size 5 cm x 5 cm, were defined at gantry angles of 0°, 90°, 180° and 270°. DVHs were then calculated on a grid with 4 mm x 4 mm x 4 mm spacing. These were then compared with DVHs calculated on a grid with 2 mm x 2 mm x 2 mm spacing, representing a DVH without grid errors. Although this test did not examine the existence of systematic errors in the calculation, it provided a useful estimate of the approximate accuracy of a typical clinical DVH.
The final test was to export the above spherical volumes of interest and dose distribution, and recalculate the DVHs using independent software [19]. Grid spacing was 4 mm x 4 mm x 5 mm for this test.
Comparisons with a second treatment planning system
In commissioning the new treatment planning system, it was important to be aware of any possible changes in calculated dose and prescribed monitor units arising from the transition between the previous system, upon which clinical experience had been built, and the new one. Hence, in-house software was written to allow CT scans and outline sets to be transferred from the Target-2 treatment planning system (Prism Microsystems, Borehamwood, UK) previously used at the Royal Marsden NHS Trust, to Pinnacle3. This facilitated the comparison of treatment plans between the two systems. By specifying identical monitor units on both treatment planning systems, both absolute dosimetry and dose distributions were compared. Target-2 used a Bentley–Milan dose calculation based directly upon measured beam data, in conjunction with CT-corrected radiological path length [20]. The collapsed cone convolution technique was used for Pinnacle3 [21, 22].
Comparison of planned and delivered dose for simulated clinical cases using thermoluminescent dosemeters
To test the overall performance of the treatment planning system, two treatment plans were created for an anthropomorphic phantom, the plans were delivered, and the planned and measured dose distributions compared. This approach was also used by McCullough and Krueger [23]. The present study was carried out for two simulated clinical cases: a pelvic (prostate) treatment and a thoracic (oesophagus) treatment. An Alderson Rando phantom was CT scanned and simulated clinical target volumes were delineated using Pinnacle3. A 3D margin of 10 mm was applied to the prostate volume and a 3D margin of 15 mm was applied to the oesophagus volume, to provide planning target volumes. The prostate case was then planned using three 6 MV fields, with gantry angles of 0°, 90° and 270° [24, 25]. The three fields were conformally shaped to the beam's eye view (BEV) of the planning target volume, with a margin of 6 mm allowed for penumbra. The fields were weighted 0.5: 0.25: 0.25 (anterior: left: right) at the isocentre, with universal wedges being applied to the lateral fields. The oesophagus case was planned using four 6 MV fields, with gantry angles of 0°, 100°, 180° and 260°. The fields were rectangular, with a 6 mm margin being allowed between the field edge and the BEV of the planning target volume [26]. Straight-edged blocks were used on the corners of the posterior oblique fields to shield the spinal cord. The fields were weighted 0.3: 0.2: 0.3: 0.2 (anterior: left posterior oblique: posterior: right posterior oblique), and no wedges were used.
Lithium fluoride thermoluminescent dosemeters (TLD) were used to verify the planned dose distributions. The TLDs were individually calibrated, using the method described by Mayles et al [27]. By delivering a prescribed dose of 1 Gy, the TLD dose–response relationship could be taken as linear. The measured doses were corrected according to the output of the linear accelerator at the time of measurement, relative to its reference output.
Comparison of planned and delivered dose for simulated clinical cases using ionization chambers (national trial comparisons)
Our centre participated in the quality assurance procedures for the START (STAndardisation of RadioTherapy) breast trial [28, 29] and Medical Research Council RT-01 prostate dose escalation trial [30]. This entailed creating standard plans, as used at this centre, for each of the treatment sites, i.e. a tangential wedged plan for the START trial and a three-field MLC plan for the RT-01 trial. These plans were then delivered to anthropomorphic phantoms and the dose measured using ionization chambers.
Quality assurance of digitally reconstructed radiographs
Attention was given to digitally reconstructed radiographs (DRRs) since these were to be used as the standard for verifying patient position. A simulated cuboidal water phantom with an off-centre cuboidal cavity of lung-equivalent density was digitized into the treatment planning system, together with planning target volume (PTV) and heterogeneity outlines. The principal dimensions of DRRs and BEVs of the phantom, with respect to the central axis of an applied field, were then measured and compared with manually calculated dimensions. The DRRs were then compared with BEVs for cardinal beam angles, together with several less regular angles. DRRs were also evaluated at several source-to-film distances. DRRs for a conformally blocked brain plan and an MLC prostate plan were also compared with corresponding BEVs.
|
Results |
|---|
|
TopAbstractIntroductionMethods and materialsResultsDiscussionReferences |
|---|
. It can be seen that the errors are less than the measurement noise for field sizes up to 20 cm x 20 cm. Even for field size 40 cm x 40 cm, the largest difference is just over 1% at depths less than 10 cm. For wedges, the difference increases to 2% at the maximum field size of 30 cm x 40 cm. The differences between ionization chamber and diode are larger in the buildup region, possibly due to geometrical and positioning discrepancies.
|
, where the energy bins of the photons are equally spaced at 0.1 MeV. However, the spectra used in Pinnacle3 are based upon a discrete set of kernels, which are not equally spaced in energy. If the values generated by Monte Carlo simulation were to be simply entered into Pinnacle3 without adjustment, the effect would be that at the higher energies, where there are fewer bins, there would appear to be fewer photons than there really were. The spectra from Monte Carlo simulation must therefore be rebinned before use, and these rebinned spectra are shown in Figure 2b. The discontinuities in the rebinned spectra occur because the spectrum in Figure 2a represents the number of photons per 0.1 MeV energy range, whereas the spectrum in Figure 2b represents the total number of photons in each energy bin. In Pinnacle3, the energy bins are 0.1 MeV wide up to 0.6 MeV, then 0.2 MeV wide up to 1.0 MeV, then 0.25 MeV wide up to 1.5 MeV, etc. Thus, each time the bin width increases, the number of photons throughout that increased energy range increases accordingly. Hence, the spectrum in Figure 2b is similar to that in Figure 2a up to 1 MeV (apart from rescaling), but then rises considerably as the curve is weighted by the increasing energy bin width.
|
, the primary fluence as simulated by Monte Carlo methods is shown for field sizes 4 cm x 4 cm, 10 cm x 10 cm and 40 cm x 40 cm. For the smaller field sizes, the fluence increase is very small, whereas for the 40 cm x 40 cm field size, the fluence increase as found by linear fit is 0.018 cm-1. However, the fluence increase which must be specified within Pinnacle3 to fit the measured dose profiles is only 0.004 cm-1 for the large field size. This difference must reflect the use of an empirical accelerator model within Pinnacle3. The model of beam softening only allows the beam energy to vary as a linear function of the angular distance from the central axis. However, as can be seen in the Monte Carlo simulations of the off-axis softening, the average energy does approximately reduce linearly (Figure 4a).
|
|
, the electron contamination, as simulated by a Monte Carlo method, is shown with a curve fit for input into Pinnacle3, i.e. the Monte Carlo simulated curve has been fitted with the parameters available in Pinnacle3 at depths below the maximum depth of electron contamination. Appropriate values of surface fraction (SF) and depth fraction (DF) have then been applied to create the linear part of the fitted curve near to the surface (less than 0.5 cm). However, in Pinnacle3, the electron contamination at the surface required to account for the difference between photon dose and measured dose is approximately 25% of the maximum dose. This is about 10 times higher than that predicted by Monte Carlo simulation. It is therefore evident that the electron contamination parameters within Pinnacle3 account for other effects beside true electron contamination. Hence, in using the result of Figure 4b, the magnitude of the electron contamination has been manipulated to fit the measured data.
Beam model accuracy
Using the comparison tools supplied with Pinnacle3, most dose distributions calculated using the Monte Carlo simulated and auto-modelled beam parameters match the measured data to within ±2%, with slightly larger discrepancies of up to ±5% observed at the peak of the dose profile for wedges. This is in accord with TG53 recommendations [3]. Absolute output measurements carried out for a range of field sizes at SSDs between 90 cm and 120 cm all show agreement with calculated values to within 2%. In addition, absolute measurements performed at SSD 441 cm, to evaluate the possibility of using Pinnacle3 for total body irradiation at extended SSD, show an agreement with calculations still within 2%.
Small-field beam model
Computed dose distributions agree with measurements to within (and in most cases significantly better than) the TG53 example acceptability criteria. Central axis depth doses for all beams (including blocked and wedged) agree with diode measurements to within 10%, 3% and 1% beyond depths of 1.5 mm, 5 mm and 10 mm, respectively. For unblocked fields, computed profiles agree with measurements to within ±1% in the in-field region, a 1–1.5% discrepancy for up to 5% of the open in-field profile width being observed in about 10% of the measurements at 20 cm depth. The corresponding discrepancy in approximately 30% of 20 cm deep, unblocked wedge profiles is 1–2% for up to 25% of the in-field profile length. Greater importance has been given to depths less than 20 cm in the modelling process as the 20 cm depth has less relevance in brain treatments. The maximum discrepancy in open beams is 1.5% while the peaks of unblocked wedge profiles show a maximum 3% disagreement. Figure 5 shows a computed planar dose distribution from an irregular block aperture compared with that measured using film. The block shape has been digitized from the 50% isodose contour on the film. Having matched the 50% isodoses, the 10%, 20%, 80% and 95% isodoses agree in most places to within 0.5 mm, the maximum difference being approximately 1.0 mm. In general, isodose lines in computed planar dose distributions for individual blocked fields are within 1 mm of film measurements. The agreement between calculated and measured blocked/unblocked dose ratio at the isocentre is ±0.4% or better for the range of blocks tested.
|
for 5 cm x 5 cm 10 MV fields delivered to layered inhomogeneous materials at gantry angle 5°. The depth dose results for lung (Figure 6a) show good agreement between Pinnacle3 and measurements both in and beyond the lung region. The treatment planning system takes into account the loss of scatter in the low density inhomogeneity (correction factor less than unity) and correctly predicts the increased dose (correction factor greater than unity) beyond the inhomogeneity due to reduced equivalent path length. The results for bone (Figure 6b,c) show poorer agreement in the inhomogeneity itself. In particular, Pinnacle3 does not seem to take account of the increased scatter in the bone, and this is particularly noticeable in the more dense bone (Figure 6c). However, extended regions of such dense bone are not often seen in clinical practice. Beyond the bone, which is more clinically applicable, the agreement between Pinnacle3 and measurements is good, with the planning system correctly accounting for the increased equivalent path length (correction factor less than unity). Figure 6d shows the inhomogeneity correction factor at different distances beyond lung for both 6 MV and 10 MV beams at gantry angle 0°. These results show that the differences between calculated and measured values are less than 2%.
|
for 5 cm x 5 cm fields delivered to the CT phantom. It can be seen that the measured dose agrees with the calculated dose to within 2.5%. These results are representative of other similar comparisons.
|
compares measured and calculated dose for oblique MLC fields delivered to a square phantom. The measured and calculated doses are found to agree to within 0.5%. Table 2 shows measured and calculated dose for irregular fields whose shapes were comparable with those described in TG53 [3]. Figure 8 shows the correspondence of calculated and measured isodoses for a diagonal MLC field. The agreement of the 50%, 80% and 95% isodoses is within 1 mm. A larger discrepancy occurs at lower doses, due to the low dose gradient. In this region, the treatment planning system predicts a lower background radiation beneath the MLC leaves than is measured, the difference being around 3%. Since the measured isodoses are similar to those planned, MLC fields can be fitted to the planning target volume with confidence. Other similar comparisons show comparable results.
|
|
|
. The centrally positioned volume receives eight distinct doses, each occupying approximately 12.5% of the volume of the cubic structure, as expected. A small volume of the structure receives doses in between the discrete levels of 10 Gy, 20 Gy, etc. i.e. slight gradient on the DVH between these doses, probably due to interpolation of the dose distribution within the treatment planning system. Absence of dose between 40 Gy and 50 Gy, i.e. DVH horizontal between 40 Gy and 50 Gy, suggests that this interpolation occurs only between doses on a given CT slice, not between. This is because regions with 10 Gy, 20 Gy, 30 Gy and 40 Gy are located inferiorly within the cubic structure, and regions receiving 50 Gy, 60 Gy, 70 Gy and 80 Gy are located superiorly. Interpolation between doses on different CT slices would therefore yield values between 40 Gy and 50 Gy.
|
), corresponding to either a half or a whole voxel, the DVHs are modified by up to 2%, the precise details being governed by the handling of the voxels within Pinnacle3. This test therefore gives confidence that the DVH calculation, although inevitably affected by the finite size of the voxels, provides an accurate result.
Figure 9b shows the comparison of DVHs using 4 mm x 4 mm x 4 mm grid spacing with those using 2 mm x 2 mm x 2 mm. Taking the DVH produced with the finer resolution to be the standard, this figure shows that errors of up to 5% dose or volume can occur. However, the previous test indicates that this is simply due to the effect of finite grid spacing, rather than any form of error.
Finally, Figure 9c gives the comparison of DVHs produced by Pinnacle3 and by an independent program. Again, differences of up to 5% dose and volume are observed, due to the finite size of the voxels, and probable differences in handling of the voxels between the two independent calculations.
Comparisons with a second treatment planning system
Table 3 shows a summary of isocentric doses calculated by Target-2 and Pinnacle3 for a range of tumour sites and treatment techniques, occurring in a range of patients. It can be seen that the doses are generally within 2%, but with a few excursions beyond this. The breast treatments differ most between the two planning systems, with Pinnacle3 always calculating a lower dose. This is due to Pinnacle3 taking into account the lack of surrounding tissue in its convolution dose calculation, whereas Target-2 simply applies a pixel-by-pixel equivalent path length correction to the dose calculation point. No changes to clinical prescriptions have been made as a result of these data.
|
shows the results of the pelvic treatment verification using TLD. The mean difference between the planned and delivered doses is 0.1±3.0% (1 standard deviation) of the isocentric dose, confirming that there is no systematic difference. The largest single difference is 8%, although this must be interpreted in relation to the accuracy of the TLDs. A 2% standard deviation is observed in the calibration TLDs, after correction for individual chip sensitivity.
|
. The mean difference between the planned and delivered doses is 0.7±2.2%, and the largest individual difference is again 8%, although most differences are significantly less than this.
Comparison of planned and delivered dose for simulated clinical cases using ionization chambers (national trial comparisons)
The dose measured at any point within the START trial phantom is always within 2.5% of the dose predicted by Pinnacle3 (Figure 11a). The largest errors are seen near the inhomogeneous lung insert. However, these results are very acceptable. Note that these measurements are included with those for other centres in reference [29].
|
shows the differences between measured and calculated doses within the RT-01 quality assurance phantom. In this case, the differences are greater, but many of the points of interest lie within the region of high dose gradient. These differences are therefore almost certainly due to small positioning errors when irradiating the phantom.
Quality assurance of digitally reconstructed radiographs
All evaluated BEVs and DRRs agree with manual calculations to within ±1 mm. The DRRs also agree with the BEVs to within ±1 mm. This aspect of the treatment planning system is therefore found to be acceptable.
|
Discussion |
|---|
|
TopAbstractIntroductionMethods and materialsResultsDiscussionReferences |
|---|
In addition to standard commissioning procedures, it has been possible to carry out further work that provides confidence in the treatment planning system. This includes (1) generation of beam parameters using Monte Carlo data, (2) use of auto-modelling facilities within Pinnacle3, (3) testing of DVHs using precise tests and independent programs, and (4) comparison of Pinnacle3 against a well established treatment planning system. It has also been possible to include results of standard tests carried out as part of national trials. This work complements the procedures given in TG53, providing additional confidence in the use of Pinnacle3.
The use of Monte Carlo simulation significantly assists in generating an accurate model within Pinnacle3. However, there are still several parameters in Pinnacle3 that are a combination of physical factors and parameters to make the model fit the measured data. It would be desirable to have a physical basis for all the parameters used in the Pinnacle3 planning system, although this may not be realistically achievable.
The auto-modelling utility provided with Pinnacle3 has been found to be very useful. This allows the user to generate beam models conveniently, and to an accuracy with which they are satisfied. Although the final models are somewhat different to those produced through the application of Monte Carlo methods, the accuracy is comparable.
For each linear accelerator and beam energy, at least two beam models have been used: one for the open fields and one for the wedged fields. For two machines, four models were used for the open fields, corresponding to field sizes of 3 cm x 3 cm, 10 cm x 10 cm, 20 cm x 20 cm and 40 cm x 40 cm. This has provided improved accuracy over the range of field sizes. Most of the models are for an SSD of 90 cm. This is a sensible arrangement because this SSD is fairly typical for a variety of clinical isocentric treatments. Accuracy of dose at other SSDs then relies on the ability of Pinnacle3's machine model to extrapolate accurately. We have verified the doses calculated at other SSDs for one linear accelerator and found the result to be acceptable. In general, the accuracy of the beam model is slightly variable over the four accelerators modelled. This experience has also been reported by Kermode et al [31] when commissioning a different treatment planning system. This probably represents the limitation of the beam model in representing the geometry and output of the linear accelerator.
Stereotactically guided conformal radiotherapy of small brain tumours demands a highly accurate dose calculation model to match the steps taken in the other processes during the design and delivery of a high-precision treatment. The fact that the effort in modelling the small-field machine data could be concentrated on a limited set of treatment parameters (limited field size range and single energy and method of field shaping) has meant that a model that is clinically acceptable for such treatments has been achieved in a practical time scale.
In general, the TPS has been commissioned to within the tolerances specified by TG53 [3]. However, even using such a sophisticated planning system as Pinnacle3, there are inevitably small regions of dose profiles that lie outside of the recommended tolerance. Complete fulfilment of all criteria within TG53 would require an impractical length of time for commissioning. Venselaar et al [5] also made the same conclusion and introduced the concept of confidence interval, in which 6.5% of dose test results were allowed to lie outside of this range. Although this criterion has not been followed in detail in the present commissioning, the results approximately satisfy it. One area where dose accuracy could be improved is in the build-up region. Here, in order to achieve an accurate prediction of dose, the maximum depth of electron contamination must be set to an unrealistically high value, compared with both Monte Carlo simulations and other measurements [32]. The parameters within Pinnacle3 evidently include other effects besides electron contamination. Dose profiles for wedged fields could also be improved. The current Pinnacle3 model assumes a single axial profile for the wedge shape, whereas in reality, the wedge is a multifaceted device. Hence, part of the wedge profile is modelled as off-axis fluence increase. This may limit the accuracy of wedged field calculations.
Assessment of 3D treatment plans is increasingly based upon DVHs and confidence in their accuracy is therefore important. However, checking the accuracy of a DVH is difficult due to the difficulty in determining the expected result, when the histogram is based upon a continuous distribution of dose. To date, the only practical method has been to choose a structure whose shape is constant in the superior–inferior direction, and then to apply fields that are long in the superior–inferior direction, and therefore approximately constant along the length of the structure, thereby reducing the problem to a two-dimensional one. Volumes and doses can then be calculated by measuring regions between isodose lines [33]. However, this is not very accurate due to the assumptions made regarding the dose distribution along the length of the structure and due to the difficulty of measuring isodose lines. Results are therefore given of more detailed examinations intended to assess accurately the performance of the DVH algorithm.
The comparison with the well-established Target-2 treatment planning system has revealed no significant systematic differences between isocentric doses for the two systems. Since the Target-2 system is based upon beam data measured on a previous occasion, the correspondence between the doses calculated by both systems gives confidence in the beam data and calculation results in the present commissioning. The lower isocentric doses predicted by Pinnacle3 for breast plans are taken to be a reflection of the superior accuracy of its scatter calculation, rather than inaccuracy, which is also confirmed with the START measurements in this paper and measurements reported elsewhere [34]. The comparison of Pinnacle3 with results obtained from TLD measurements in a humanoid phantom also provide additional confidence. Although differences of up to 8% between calculated and measured dose are observed in specific regions, the differences are generally much less than this, and there is no systematic difference. Similarly, the comparisons of Pinnacle3 with measurements in the START and RT-01 phantoms using ionization chambers show excellent agreement.
This paper describes the clinical commissioning of the Pinnacle3 planning system for traditional and conformal techniques. Further work is underway to commission the planning system for intensity-modulated radiotherapy using multiple static fields. In this case, the accuracy of the inverse planning module, the suitability of field segments, and the dosimetric accuracy of the computed dose distributions relative to measurements in realistic phantoms, must all be taken into account. The commissioning described in this paper is expected to provide a foundation for this work.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Received for publication April 23, 2002. Revision received August 30, 2002. Accepted for publication September 26, 2002.
|
References |
|---|
|
TopAbstractIntroductionMethods and materialsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  M Mazonakis, J Damilakis, H Varveris, and N Gourtsoyiannis Therapeutic external irradiation in women of reproductive age: risk estimation of hereditary effects Br. J. Radiol., October 1, 2004; 77(922): 847 - 850. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||
, where Fd (d, s) is the electron contamination factor as a function of depth d and field size s, Fs (s) is the dependence of electron contamination on field size, k is a constant, and dme is the maximum depth at which electron contamination occurs. SF is the ratio of electron contamination at the surface to that described by the above equation at d=0. Fs (s)=0.06, SF=0.6, k=1.3, and dme=2.2 cm. At depths less than DF.dme, where DF=0.2, the curve is linear, such that the surface dose is Fs (s).
