If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Medical Physics and Radiation Protection Department, Catalan Institute of Oncology Girona, Girona, SpainDepartment of Medical Sciences, University of Girona, Girona, Spain
Planning with dose-to-medium as with dose-to-water can impact clinical outcome and robustness.
•
Dose homogeneity should not be requested if there are media with different responses.
•
Uniform irradiation should be pursued.
•
Avoid dose-to-medium sensitivity to composition when planning and evaluating.
Abstract
Background and purpose
Radiotherapy planning considerations were developed for the previous calculation algorithms yielding dose to water-in-water (Dw,w). Advanced algorithms improve accuracy, but their dose values in terms of dose to medium-in-medium (Dm,m) depend on the medium considered. This work aimed to show how mimicking Dw,w planning with Dm,m can introduce new issues.
Materials and methods
A head and neck case involving bone and metal heterogeneities outside the CTV was considered. Two different commercial algorithms were used to obtain Dm,m and Dw,w distributions. First, a plan was optimised to irradiate the PTV uniformly and get a homogeneous Dw,w distribution. Second, another plan was optimised to achieve homogeneous Dm,m. Both plans were calculated with Dw,w and Dm,m, and the differences between their dose distributions, clinical impact, and robustness were evaluated.
Results
Uniform irradiation produced Dm,m cold spots in bone (-4%) and implants (-10%). Uniform Dm,m compensated them by increasing fluence but, when recalculated in Dw,w, the fluence compensations produced higher doses that affected homogeneity. Additionally, doses were 1% higher for the target, and + 4% for the mandible, thus increasing toxicity risk. Robustness was impaired when increased fluence regions and heterogeneities mismatched.
Conclusion
Planning with Dm,m as with Dw,w can impact clinical outcome and impair robustness. In optimisation, uniform irradiation instead of homogeneous Dm,m distributions should be pursued when media with different Dm,m responses are involved. However, this requires adapting evaluation criteria or avoiding medium effects. Regardless of the approach, there can be systematic differences in dose prescription and constraints.
]. This objective was established to ultimately deliver a uniform dose to the CTV. As CTV composition is usually homogeneous, this is achieved by irradiating it with a uniform photon energy fluence, but this fluence must be extended within the PTV to cover all CTV contouring uncertainties and positions throughout the treatment. Consequently, the resulting goal is to irradiate the PTV uniformly, which translated into achieving a homogeneous dose in the PTV.
This approach was developed for the previous-generation dose calculation algorithms that considered all tissues and materials to be water of different densities. However, more accurate algorithms must distinguish between media. Consequently, apart from their inherent accuracies, some of the discrepancies among algorithms are due to how they handle the different media. This results in three distinct dose quantities that can be described as Ddm,tm, where dm is the medium of the deposition voxel, tm is the medium of radiation transport, and both can be either w for water or m for medium.
Treatment planning system (TPS) algorithms can be roughly classified according to their associated dose quantity [
Kry SF, Feygelman V, Balter P, Knöös T, Ma C ‐M. C, Snyder M, et al. AAPM Task Group 329: Reference dose specification for dose calculations: dose‐to‐water or dose‐to‐muscle? Med Phys 2019:mp.13995. 10.1002/mp.13995.
]. Dose to water-in-water (Dw,w) is reported by convolution/superposition (C/S) algorithms using pencil beam kernels and some using point kernels. Dose to medium-in-medium (Dm,m) is reported by Monte Carlo (MC), grid-based Boltzmann solvers (GBBS), and some implementations of the Collapsed Cone point kernel C/S algorithm. Actually, the doses of point kernel C/S algorithms are a mix of Dm,m and Dw,w, being closer to one or the other depending on the implementation. Lastly, dose to water-in-medium (Dw,m) can also be reported by MC and GBBS advanced algorithms. These advanced algorithms can generally be configured to report Dm,m or Dw,m, although some MC implementations define all materials as water and yield Dw,w. Each quantity has its advocates [
Kry SF, Feygelman V, Balter P, Knöös T, Ma C ‐M. C, Snyder M, et al. AAPM Task Group 329: Reference dose specification for dose calculations: dose‐to‐water or dose‐to‐muscle? Med Phys 2019:mp.13995. 10.1002/mp.13995.
ICRU. ICRU Report 91, Prescribing, Recording, and Reporting of Stereotactic Treatments with Small Photon Beams. Bethesda, MD: 2017. 10.1007/s00066-018-1416-x.
Benchmark Credentialing Results for NRG-BR001: The First National Cancer Institute-Sponsored Trial of Stereotactic Body Radiation Therapy for Multiple Metastases.
The approaches followed to calculate Dw,w and Dm,m yield different doses in different media because photon and electron interactions depend on atomic composition [
]. This affects radiation transport and energy deposition, introducing discrepancies within materials and at material interfaces. For soft tissues, Dw,w and Dm,m values are around 1% similar for a range of beam energies and depths [
Kry SF, Feygelman V, Balter P, Knöös T, Ma C ‐M. C, Snyder M, et al. AAPM Task Group 329: Reference dose specification for dose calculations: dose‐to‐water or dose‐to‐muscle? Med Phys 2019:mp.13995. 10.1002/mp.13995.
]. If several materials are involved, build-up and build-down regions exist at material interfaces that can only be correctly modelled by advanced algorithms [
A new dose quantity for evaluation and optimisation of MV photon dose distributions when using advanced algorithms: proof of concept and potential applications.
Most TPSs use simple Dw,w algorithms in plan optimisation for speed purposes. Still, computation improvements allow the introduction of more accurate algorithms, being feasible to optimise plans for Dm,m in some systems. This is advantageous if the final plan is also calculated in Dm,m because the whole process is consistent, and the results are more likely to fulfil the objectives [
]. However, the differential behaviour of Dw,w and Dm,m can affect the goal of delivering a uniform dose to the PTV to assure a homogeneous CTV dose in all circumstances. This is illustrated in Fig. 1, where a uniform photon energy fluence irradiates a PTV encompassing a homogeneous CTV and part of a bone heterogeneity. Assuming charged particle equilibrium exists, the resulting dose distribution is also uniform if a Dw,w algorithm is considered since atomic composition is constant and dose is locally independent of density according to Fano’s theorem [
]. However, if a Dm,m algorithm is used, the dose response in the heterogeneity differs and the dose distribution is inhomogeneous. In a case like this, achieving a uniform Dm,m distribution during optimisation would require introducing local fluence compensations in the heterogeneity, thus deviating from uniform fluence.
Fig. 1Dw,w and Dm,m distributions when a PTV encompassing a homogeneous CTV and bone (or other atomic composition heterogeneity) is irradiated uniformly.
How to proceed in this situation when using Dm,m needs to be clarified. There are two options: irradiate uniformly, or request homogeneous Dm,m distributions to mimic previous planning based on Dw,w. The first option keeps clinical practice but implies accepting inhomogeneous Dm,m distributions that are difficult to evaluate. The second option modifies patient irradiation and deviates from previous clinical practice, which can have clinical consequences in all treatment scenarios.
This work aimed to demonstrate that mimicking Dw,w planning with Dm,m to achieve homogeneous Dm,m distributions can introduce new clinical and robustness issues to consider if atomic composition heterogeneities are involved.
2. Materials and methods
A head and neck case involving air, bone, teeth, and implants outside the CTV and partially inside the PTV was considered for illustration. No approval was required by our institutional ethics committee. Simultaneous integrated boost was delivered to three PTVs (54, 60, and 70 Gy) in 33 fractions. For clarity, we focused on the CTV70Gy, the PTV70Gy (5 mm margin), and the mandible (metal implants excluded) as organ-at-risk (OAR).
The planning process was performed in the Eclipse v15.6 TPS (Varian Medical Systems, Palo Alto, CA, USA). The AAA v15.6.05C/S algorithm and the Acuros XB (AXB) v15.6.05 GBBS algorithm were used to obtain the Dw,w and the Dm,m distributions, respectively, with 1 mm voxel size.
Depending on the algorithm, the medium characterisation of each CT image voxel is performed differently. AAA retrieves its electron density using a CT number-to-electron density table. AXB requires its mass density and atomic composition. The mass density is derived analogously, and is then linked to a material with a well-known composition specified in the AcurosXB-13.5 material library [
]. The library comprises several tissues for automatic assignment, each associated with a density range. Adjacent tissues present regions of overlapping densities which are modelled to change linearly from one tissue to another, thus providing a mixture that enables smooth transitions. The library also includes materials that users can assign manually.
Fig. 2 presents the case and the composition of each volume considered by AXB, provided by in-house software that emulated the AXB medium characterisation process.
Fig. 2Head and neck case used in this work. For each structure, percentages of each pure medium as considered in the Acuros XB v13.5 material table (densities in the overlapping regions decomposed into their pure components).
2.1 Plan with uniform Dw,w and plan with uniform Dm,m in the PTV70Gy
We optimised two plans with volumetric modulated arc therapy (VMAT) for a 6 MV photon beam from a Varian TrueBeam accelerator (Varian Medical Systems).
The first plan was optimised to irradiate the PTV uniformly. This could be done directly in Eclipse because it used a dedicated simple algorithm (MRDC) yielding Dw,w to speed up the optimisation [
] and, consequently, the resulting plan used a uniform photon energy fluence to obtain homogeneous MRDC Dw,w.
The second plan was optimised to mimic previous Dw,w distributions and achieve uniform Dm,m in the PTV. This was done using Eclipse's “intermediate dose” option during optimisation [
]. This option allowed performing a more accurate calculation in one of the optimisation steps, computed the discrepancies with the distribution calculated by the internal Dw,w algorithm at this step, and incorporated the voxel-by-voxel differences as penalties into the optimiser's cost function. Thus, if the intermediate dose calculation was performed with AXB Dm,m and PTV dose homogeneity was requested in the objectives, the resulting optimised plan presented a homogeneous AXB Dm,m in the PTV.
Both plans were calculated with AXB Dm,m and normalised to the PTV70Gy median dose. They were also calculated with AAA to analyse the differences with Dw,w-based clinical practice.
2.2 Plan analysis
The dose distributions were quantitatively evaluated in the nominal scenario for the target volumes using dose-volume-histograms (DVH) and the ICRU 83 reporting parameters [
]: near-minimum dose D98%, median dose D50%, near-maximum dose D2%, and homogeneity index HI=(D2%-D98%)/D50%. A slice-by-slice visual inspection was also performed to investigate spatial correlations between dose and medium heterogeneities. These evaluations were done with reference to the detailed composition information provided in Fig. 2.
For the second plan with uniform Dm,m, we evaluated the clinical impact associated with its Dm,m and Dw,w distributions in the nominal scenario, and plan robustness [
]. Plan complexity was not considered as it is particularly useful when comparing plans, but there is only one in this case.
The clinical impact was evaluated in the nominal scenario by comparing target response and mandible toxicity predictions made by Dm,m with those made by Dw,w -for which most clinical experience was established-. For the target, we compared the DVHs for the CTV70Gy. For the mandible, we compared the normal tissue complication probabilities (NTCP) estimated using the van Dijk et al. model for osteoradionecrosis [
Normal Tissue Complication Probability (NTCP) Prediction Model for Osteoradionecrosis of the Mandible in Patients With Head and Neck Cancer After Radiation Therapy: Large-Scale Observational Cohort.
The prevalence and risk factors associated with osteoradionecrosis of the jaw in oral and oropharyngeal cancer patients treated with intensity-modulated radiation therapy (IMRT): The Memorial Sloan Kettering Cancer Center experience.
]. The additional treatment scenarios consisted of rigid translational setup errors in both directions of the three Cartesian axes, and towards the nearest main heterogeneity (worst-case scenario). They were simulated by shifting 5 mm (CTV-PTV margin) the plan's isocenter, and the dose distributions were calculated in Dm,m. Plan robustness was quantified by the variations across scenarios of the dosimetric parameters mentioned above for the CTV70Gy.
Additionally, we analysed the photon energy fluence distributions to facilitate the interpretation of the results. Their accurate computation requires MC calculations, but this is complex and involves external tools. In this study, we were more interested in unveiling fluence variations and their spatial correlation with medium heterogeneities than in accuracy. For this reason, we used the Dw,w distributions as surrogates since, as their dose values are not sensitive to atomic composition heterogeneities, any Dw,w inhomogeneity would correspond to actual fluence changes.
3. Results
Fig. 3 and Table 1 show the dose distributions, the DVHs, and the dosimetric parameters for the plans optimised to achieve homogeneous Dw,w and homogenous Dm,m in the PTV, both reported in Dw,w and Dm,m.
Fig. 3Dm,m and Dw,w distributions, and DVHs for the target volumes (CTV70Gy orange, PTV70Gy red) for the plans with homogeneous Dw,w (a)(b) and homogeneous Dm,m (c)(d) in the PTV70Gy. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
The optimisation resulted in a homogeneous Dw,w distribution in the PTV70Gy (Fig. 3(b)). If calculated in Dm,m (Fig. 3(a)), the dose in the CTV70Gy -consisting almost exclusively of muscle (Fig. 2)- was also homogeneous. However, the PTV70Gy presented cold regions of around −4% and −10% matching bone and metal implants, respectively. Due to the proportion of these materials in the volume, the DVH was distorted and some dosimetric parameters were affected, with D98% well below 95% -and beyond classical acceptability criteria- and a low HI value.
3.2 Plan with uniform Dm,m in the PTV
In terms of Dm,m (Fig. 3(c)), the dose inhomogeneities of the previous plan were corrected and all dosimetric parameters were within planning objectives. Nevertheless, hot regions appeared if the dose distribution was calculated in Dw,w (Fig. 3(d)), increasing D2% and inhomogeneity above acceptability criteria. Visual inspection showed that these higher Dw,w doses matched the location and value (although in the opposite direction) of the previous plan's lower Dm,m regions. Interpreting the Dw,w distribution as a surrogate of actual photon fluence energy distribution, this means that fluence increases locally in bone and metal to compensate for the lower Dm,m values and achieve Dm,m homogeneity.
The clinical impact in the nominal scenario is shown in Fig. 4. The CTV70Gy received the prescribed dose homogeneously for Dm,m, but the corresponding Dw,w values were around 1% higher systematically. For the mandible (metal implants excluded), Dm,m values were about 4% lower than those of Dw,w and all the dosimetric parameters associated with toxicity appeared to be lower. Taking as reference the values calculated with Dw,w, the toxicity risk was significantly higher than that predicted with Dm,m, and the NTCP was underestimated by up to 6%.
Fig. 4Clinical impact for the CTV70Gy (a) and the mandible (b) in the nominal scenario. Solid and dashed lines in the DVHs correspond to Dm,m and Dw,w distributions, respectively. For the mandible, the parameters associated with toxicity are presented and compared for each dose quantity.
Lastly, CTV70Gy robustness against setup errors is presented in Fig. 5. D2% variations from the nominal scenario generally exceeded 2% and were as high as 6.6%. Assessing the scenarios slice by slice, we found that this lack of robustness was due to positional mismatches between the atomic composition heterogeneities and the corresponding fluence increases introduced during optimisation.
Fig. 5CTV70Gy robustness against setup errors. Metal implants contoured in pink. The DVH combines the DVHs of the nominal (solid) and shifted (dashed) scenarios. For each dosimetric parameter, the percentage differences between the nominal scenario and the scenarios where presented its minimum/maximum values are shown. Red highlights differences higher than 2%. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Trying to mimic with Dm,m the homogeneity of previous Dw,w distributions seems natural. However, this should be done cautiously if the PTV composition is heterogeneous. The study was performed in Eclipse, but the conclusions should also apply to any TPS capable of optimising Dm,m distributions obtained with MC, GBBS, or, to a lesser extent, Collapsed Cone Dm,m implementations. To our knowledge, only Eclipse TPS currently allows this, but the increasing adoption of advanced algorithms and computational improvements will make it feasible in more TPSs in the future, as exemplified in Feygelman et al. [
The underlying cause behind the new issues is the dependence of Dm,m values on medium atomic composition. If the PTV contains materials with different dose responses and homogeneity is requested, local fluence variations must be introduced to compensate for the different dose values in these media. The influence of material interface effects can be ignored as a first approximation since their extent is limited to a few mm and are smoothed out for typical voxel sizes [
]. Furthermore, the percentage of voxels affected in clinical practice is generally low, and abrupt transitions between materials are rare due to partial volume effects in the CT images (and in this study also because of the gradual mixture for nearby densities).
In any case, and as discussed in the literature, these compensations should be avoided [
]. On the one hand, they deviate from previous practice where fluence was homogeneous. Patients are irradiated differently, usually with higher doses as fluence needs to be increased in most media involved in radiotherapy planning. The clinical impact of these higher doses depends on whether they are located in the CTV or an OAR partially included in the PTV, thus increasing tumour control probability or toxicity risk, respectively. On the other hand, robustness is impaired if the fluence variations and their corresponding heterogeneities mismatch (Fig. 5). The impact of these issues depends on the type and relative proportion of atomic composition heterogeneities. In this work, we used an illustrative case with a relatively high volume of non-water-like media in the PTV, but in clinical routine there is a wide variability in the proportions and their influence [
]. Nevertheless, all the problems described can occur, and their effects can be significant depending on the case. It is worth noting that they can be overlooked if plan analysis is performed solely based on the Dm,m distribution, as is the case for the two works that study plans specifically optimised for Dm,m algorithms [
]. More aspects should be analysed if this planning strategy is adopted.
As has been argued, plans with uniform irradiation are preferable for clinical and robustness reasons. However, they are neither straightforward to obtain nor to evaluate with Dm,m. Plan evaluation can be affected as Dm,m inhomogeneities can alter the values of some dosimetric parameters beyond acceptability criteria. This way, it is hard to distinguish if a plan does not fulfil the objectives because of these effects or because it is not optimal. This problem also disturbs plan optimisation since it is driven by the evaluation of the dose distribution within each iteration using dosimetric instead of fluence-related parameters. As a result, the optimiser inevitably introduces the fluence compensations we try to avoid.
There are several approaches to solve this, although neither is optimal. Theoretically, robust optimisation would be the preferred option. Dispenses with PTV and should produce a more homogeneous fluence when necessary to reach a compromise between robustness and dose objectives. However, it is not available for photons in most TPSs, it increases optimisation time, and it is not well-established how robust evaluation should be performed.
Another possibility could be adapting ICRU criteria (minimum target dose, homogeneity, etc.) and optimisation objectives. Nevertheless, it is unclear how this could be done depending on the media involved and their relative amounts within volumes.
The most practical solution is to avoid Dm,m sensitivity to medium composition. Various approaches can be used [
A new dose quantity for evaluation and optimisation of MV photon dose distributions when using advanced algorithms: proof of concept and potential applications.
]: override the material, optimise and/or recalculate in Dw,w, or use the dose to reference-like medium (Dref,m*). The first two options can compromise radiation transport accuracy. Dref,m* is a new quantity based on advanced algorithms that keeps radiation transport accuracy and whose values are not sensitive to medium atomic composition [
] and, although it is not yet available in TPSs, Dref,m* is simple to implement.
Regardless of the planning strategy adopted, dose prescription and clinical response prediction are affected if the reference data is based on Dw,w. Targets usually present a water-like composition, and Dm,m values are 1% lower. This difference is addressed in the AAPM TG-329 document [
Kry SF, Feygelman V, Balter P, Knöös T, Ma C ‐M. C, Snyder M, et al. AAPM Task Group 329: Reference dose specification for dose calculations: dose‐to‐water or dose‐to‐muscle? Med Phys 2019:mp.13995. 10.1002/mp.13995.
] by decreasing the doses of Dw,w algorithms (tuning the TPS configuration) to report dose-to-muscle values. If prescription doses are not rescaled accordingly, the doses received by patients are systematically 1% higher than before. For bony organs, the dose discrepancies increase up to 4% for cortical bone, and the risk of toxicity may be underestimated. Specific Dm,m constraints should be used, or Dw,w should be calculated for evaluation.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Kry SF, Feygelman V, Balter P, Knöös T, Ma C ‐M. C, Snyder M, et al. AAPM Task Group 329: Reference dose specification for dose calculations: dose‐to‐water or dose‐to‐muscle? Med Phys 2019:mp.13995. 10.1002/mp.13995.
ICRU. ICRU Report 91, Prescribing, Recording, and Reporting of Stereotactic Treatments with Small Photon Beams. Bethesda, MD: 2017. 10.1007/s00066-018-1416-x.
Benchmark Credentialing Results for NRG-BR001: The First National Cancer Institute-Sponsored Trial of Stereotactic Body Radiation Therapy for Multiple Metastases.
A new dose quantity for evaluation and optimisation of MV photon dose distributions when using advanced algorithms: proof of concept and potential applications.
Normal Tissue Complication Probability (NTCP) Prediction Model for Osteoradionecrosis of the Mandible in Patients With Head and Neck Cancer After Radiation Therapy: Large-Scale Observational Cohort.
The prevalence and risk factors associated with osteoradionecrosis of the jaw in oral and oropharyngeal cancer patients treated with intensity-modulated radiation therapy (IMRT): The Memorial Sloan Kettering Cancer Center experience.
00034-9&id=gr1.jpg){kind=link}
00034-9&id=gr2.jpg){kind=link}
00034-9&id=gr3.jpg){kind=link}
00034-9&id=gr4.jpg){kind=link}
00034-9&id=gr5.jpg){kind=link}