- Email: [email protected]
Advantages and limitations of navigation-based multicriteria optimization (MCO) for localized prostate cancer IMRT planning
Medical Dosimetry ] (2014) ]]]–]]]
Medical Dosimetry journal homepage: www.meddos.org
Advantages and limitations of navigation-based multicriteria optimization (MCO) for localized prostate cancer IMRT planning Conor K. McGarry, Ph.D.,* Rasmus Bokrantz, Ph.D.,†‡ Joe M. O’Sullivan, F.F.R. (RCSI),§║ and Alan R. Hounsell, Ph.D.*§ Radiotherapy Physics, Northern Ireland Cancer Centre, Belfast Health and Social Care Trust, Belfast, Northern Ireland, UK; †Optimization and Systems Theory, KTH Royal Institute of Technology, Stockholm, Sweden; ‡RaySearch Laboratories, Stockholm, Sweden; §Centre for Cancer Research and Cell Biology, Queen’s University Belfast, Belfast, Northern Ireland, UK; and ║Clinical Oncology, Northern Ireland Cancer Centre, Belfast Health and Social Care Trust, Belfast, Northern Ireland, UK
n
A R T I C L E I N F O
A B S T R A C T
Article history: Received 22 October 2013 Accepted 3 February 2014
Efficacy of inverse planning is becoming increasingly important for advanced radiotherapy techniques. This study’s aims were to validate multicriteria optimization (MCO) in RayStation (v2.4, RaySearch Laboratories, Sweden) against standard intensity-modulated radiation therapy (IMRT) optimization in Oncentra (v4.1, Nucletron BV, the Netherlands) and characterize dose differences due to conversion of navigated MCO plans into deliverable multileaf collimator apertures. Step-and-shoot IMRT plans were created for 10 patients with localized prostate cancer using both standard optimization and MCO. Acceptable standard IMRT plans with minimal average rectal dose were chosen for comparison with deliverable MCO plans. The trade-off was, for the MCO plans, managed through a user interface that permits continuous navigation between fluence-based plans. Navigated MCO plans were made deliverable at incremental steps along a trajectory between maximal target homogeneity and maximal rectal sparing. Dosimetric differences between navigated and deliverable MCO plans were also quantified. MCO plans, chosen as acceptable under navigated and deliverable conditions resulted in similar rectal sparing compared with standard optimization (33.7 ⫾ 1.8 Gy vs 35.5 ⫾ 4.2 Gy, p ¼ 0.117). The dose differences between navigated and deliverable MCO plans increased as higher priority was placed on rectal avoidance. If the best possible deliverable MCO was chosen, a significant reduction in rectal dose was observed in comparison with standard optimization (30.6 ⫾ 1.4 Gy vs 35.5 ⫾ 4.2 Gy, p ¼ 0.047). Improvements were, however, to some extent, at the expense of less conformal dose distributions, which resulted in significantly higher doses to the bladder for 2 of the 3 tolerance levels. In conclusion, similar IMRT plans can be created for patients with prostate cancer using MCO compared with standard optimization. Limitations exist within MCO regarding conversion of navigated plans to deliverable apertures, particularly for plans that emphasize avoidance of critical structures. Minimizing these differences would result in better quality treatments for patients with prostate cancer who were treated with radiotherapy using MCO plans. & 2014 American Association of Medical Dosimetrists.
Keywords: IMRT MCO Prostate
Background Intensity-modulated radiation therapy (IMRT) has been shown to be superior to conformal radiation therapy for patients with prostate cancer.1 Generating high-quality IMRT plans is, however,
Reprint requests to: Conor Kevin McGarry, Ph.D., Radiotherapy Physics, Northern Ireland Cancer Centre, Belfast Health and Social Care Trust, Lisburn Road, Belfast BT9 7AB, UK. Tel.: þ44 2895 048 327. E-mail: [email protected] http://dx.doi.org/10.1016/j.meddos.2014.02.002 0958-3947/ Copyright Ó 2014 American Association of Medical Dosimetrists.
often a time-consuming process because the priorities of conflicting planning goals need to be adjusted and the treatment plan reoptimized multiple times, until it is acceptable for clinical delivery. The quality of plans generated by standard methods has also been shown to depend on both experience level and time commitment of the individual treatment planner.2,3 A further difficulty is that differences between optimized and deliverable treatment plans can occur, especially when the plan requires interpretation of multileaf collimator (MLC) leaf positions.4 Developments such as direct-aperture optimization have minimized
2
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]]
these differences by including the MLC positions and associated limitations within the optimization step of creating an IMRT plan.5 Improvements in treatment planning efficacy for advanced radiotherapy techniques are becoming more important with the onset of adaptive radiotherapy. Multicriteria optimization (MCO) represents a significant step toward minimizing the number of manual iterations or recalculations during the creation of an IMRT plan.6-9 This technology utilizes a database of precomputed IMRT plans whose dose distributions are continuously combined to produce a single “navigated” treatment plan. Dose averaging over the database plans is performed according to the input from a graphical user interface that contains a set of slider bar controls that each represents the priority of a single planning objective.8,9 Navigation is feasible in real time and, thereby, permits the treatment planner to obtain immediate feedback regarding how a change of the priority of one objective correlates with the other objectives as well as how it affects the 3-dimensional dose distribution. MCO of this form has, for a number treatment sites, been shown to significantly reduce treatment planning times and simultaneously result in treatment plans that physicians judged as superior to conventionally optimized IMRT plans in blinded assessments.10,11 However, there is a paucity of data on comparisons between MCO and standard planning in terms of quantitative statistics on planned physical dose distributions for patients with prostate cancer. To complement the findings by previous authors that MCO enables planners to more easily find a suitable trade-off between objectives, this study, therefore, sought to quantify the costs and benefits of MCO with respect to dose distribution quality. This investigation is performed with respect to a suite of patients with localized prostate cancer with multiple target regions at different dose levels, as prescribed by the conventional or hypofractionated high dose IMRT trial for prostate cancer (CHHiP) protocol.12 As a secondary goal, this study also addressed the fact that a predominant number of approaches to MCO in the literature rely on databases of fluence-based treatment plans.6-9 This property can be traced to the fact that the set of database plans that form the basis for the navigated dose in MCO are computed toward forming a representation of the set of Pareto optimal treatment plans, i.e., feasible treatment plans that are nondominated in the sense that there does not exist another feasible plan that is strictly better in at least one objective and no worse in every other objective. A fluence-based representation of this set (that is subsequently called the Pareto front for short) has the advantage that there exists a one-to-one correspondence between a linear combination in dose and the corresponding linear combination in fluence. By virtue of this fact, a combination of multiple Pareto optimal fluence profiles is guaranteed to produce another feasible and near-optimal fluence-based treatment plan. This property does not extend to databases that contain deliverable treatment plans because there is no trivial way of combing multiple MLC segments into a single treatment plan with dose identical to the corresponding combination of dose distributions without increasing the total number of segments. Current research is attempting to address this issue through multicriteria direct-aperture optimization13 or segmentation during the navigation phase.14 A fluencebased representation, however, necessitates a conversion into deliverable MLC segments that can compromise plan quality. We critically assess the dose discrepancies that occur owing to this conversion and study how the magnitude of the conversion errors correlates with the spatial shape of the navigated dose distributions. Pareto fronts are not only useful as a model of the relevant treatment options, but can also be used as a form of assessment of plan quality that is not biased by subjective decisions on which
aspects of plan quality to prioritize.15 Such comparisons have previously been applied to the assessment of the effect of different planning parameters for a fixed delivery technique16 as well as assessment of differences between alternate treatment modalities.17,18 In this study, we assessed plan quality of the MCO plans for patients with prostate cancer compared with the standard IMRT plan. The accuracy of MCO plan delivery was validated. We also examined plan quality of treatment plans generated by navigated vs the corresponding deliverable MCO plans by comparison of the Pareto front spanned by target homogeneity vs average dose to the rectum.
Methods and Materials Patients Plans were created using computed tomography (CT) datasets of 10 successive patients with prostate cancer (Ethics REC ref 09/NIR02/28). Each patient had been CT scanned using a GE LightSpeed wide-bore scanner (GE Healthcare, WI) at 2.5mm slices. Target volumes were expanded and organs at risk (OARs) delineated, including the rectum, bladder, femoral heads, penile bulb, and bowel, according to the CHHiP protocol.12 Plans were created with a prescription of 60 Gy in 20 fractions, a schedule that has been used in previous planning studies.25 Dose levels to the targets (planning target volumes [PTVs] 1 through 3) and tolerances for OARs are shown in Table 1. An avoidance shell (1-cm thick) was delineated, which ensured a steep fall-off of dose from the targets to the surrounding area, as illustrated in Fig. 1. The avoidance shell was designed to not involve the rectum to ensure that target homogeneity was traded against sparing of the rectum.
Treatment planning Standard IMRT treatment plans were optimized on a single PC (HP z8000 Workstation, 2 Intel Xeon CPU X5660 2.8 GHz and 12 GB of RAM) using the direct step-and-shoot (DSS) optimization algorithm of the Oncentra treatment planning system v4.1 (Nucletron BV, Veenendal, the Netherlands). An enhanced pencil-beam (PB) dose calculation algorithm was used for final dose calculations. This dose algorithm includes modeling of the MLC type, MLC transmission, and source size. The PB algorithm was used as it has been fully validated in our department as well as other departments.19 Treatment plans were designed for delivery on a Varian 2100CD linear accelerator equipped with 120-leaf millennium MLC (Varian Medical Systems, Palo Alto, CA) via the ARIA record and verify system, using 6-MV photons. MCO plans were created on a single PC (HP EliteBook 8530w, 2 Intel Core CPU T9400 2.53 GHz and 6 GB of RAM) using the RayStation treatment planning system v2.4 (RaySearch Laboratories, Stockholm, Sweden). The database of fluence-based treatment plans that form the basis of a navigated MCO plan is in this system calculated toward minimizing the approximation error of the database representation with respect to the underlying Pareto front.20 Navigation is performed by realtime linear interpolation of dose over the database plans. A navigated treatment plan selected for conversion by the user is converted to deliverable MLC segments through DSS optimization.21 The objective of this optimization is to minimize the error between the dose-volume histogram (DVH) distributions of the navigated and the deliverable treatment plan.22 Dose for the MCO plans was computed by an algorithm that implements singular value decomposition (SVD) of PB kernels.23 An intermediate and a final dose calculation were in addition performed during DSS optimization using a collapsed-cone (CC) dose algorithm.21 The intermediate CC dose was used as background dose for subsequent SVD dose calculations. Dose calculations using SVD and CC were both performed with respect to a machine Table 1 Selection of the dosimetric quality parameters outlined in the CHHiP trial protocol CHHiP trial parameter
Constraint (%)
CHHiP trial parameter
Constraint
PTV 1 min PTV 2 min PTV 3 min Rectum v68% max Rectum v81% max Rectum v88% max Rectum v95% max Rectum v100% max
76 91 95 60 50 30 15 3
Bladder v68% max Bladder v81% max Bladder v100% max Penile bulb v68% max Penile bulb v81% max Bowel v68% max Femoral heads v68% max
50% 25% 5% 50% 10% 17 cc 50%
PTV min is the minimum dose (%) received by 99% of the volume. For critical structures, the constraint is the maximum dose received by the volume X where v(X)max is shown (e.g., X ¼ 68 for femoral heads).
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]]
3
Fig. 1. Targets outlined to receive at least 76% (PTV 1), 91% (PTV 2), and 95% (PTV 3) of prescribed dose with avoidance of the rectum shown. A shell (red), not including the rectum was included to ensure conformity. (Color version of figure is available online.) model commissioned according to experimental beam data for the linear accelerator used for this study. Standard IMRT plans, as well as MCO plans, were planned toward step-andshoot delivery with a dose rate of 400 MU/min. A 5-field class solution was used with beam angles of 351, 1001, 1801, 2601, and 3251 for both planning methods. This class solution has been developed from beam angles published previously.24 The number of segments was 50 per plan, the minimum field size was 4 cm2, and the grid size was set to 0.25 cm in all directions. All plans were assessed in terms of acceptability according to the CHHiP protocol.12 Each of the indices specified in Table 1 were also compared across all the plans. Further analysis was performed on targets in the form of homogeneity index (HI) to PTV 3 defined as D5% D95% (where D5% is the dose to 5% volume and D95% is dose to 95% volume), and conformity index calculated as the volume of the body receiving the prescribed dose divided by the corresponding PTV. The lower the conformity index, the more conformal the plan is expected to be, and the lower the HI the more uniform the dose delivered to the target.
Dose calculation validation As 2 different dose calculation engines were used between the MCO plans and the standard plans generated using Oncentra, one of the plans was tested for delivery accuracy. A deliverable plan (patient 2—generated by DSS optimization in Oncentra) was recalculated on a CT scan set of a MatriXX Evolution 2D ionization chamber array device (IBA, Schwarzenbruck, Germany), with a 6-cm thick 30 30cm2 solid water slab above and below the device, using the CC algorithm in RayStation and the PB algorithm in Oncentra. Both plans were delivered to the array using a Varian 2100CD linear accelerator, and the measured data in the coronal plane were compared with that calculated using gamma analysis with the dose and distance criteria set at 3% and 3 mm, respectively.
D95% Z 57 Gy, PTV 2 PTV 3 to satisfy D100% Z 52 Gy and D5% r 61 Gy, and PTV 2 PTV 1 to satisfy D100% Z 46 Gy and D5% r 57 Gy. A dose-limiting constraint on the global maximum dose was finally introduced at 63 Gy. The stated constraints collectively ensure that the database representation of the Pareto front is restricted to clinically relevant treatment plans. A Pareto front of target homogeneity vs average rectal dose was, after database generation, extracted by navigation over the database plans. A navigated plan that gave maximum priority to target homogeneity and that had priorities for the other goals set to ensure plan acceptability was first created. The priority to rectal sparing was subsequently increased in incremental steps (through a movement of the slider bar control associated with this objective), while adjustments were performed using the slider bars of the other objectives to retain acceptability of the created plan when possible. A series of navigation states at discrete steps along the trajectory between maximal homogeneity and maximal rectal sparing were saved as unique treatment plans. These plans were subsequently assessed for acceptability analogous to the assessment of the conventionally optimized plans. Each navigated plan was converted into deliverable MLC segments and final dose computed by CC. To facilitate a fair comparison with the conventionally optimized plans, the deliverable plans were transferred to Oncentra where the final dose was recalculated using the PB dose algorithm, and the plans again assessed for acceptability. To analyze changes between navigated and deliverable plans, plots were created of average rectal dose against homogeneity of PTV, defined using the HI. These were considered to be 2 factors that could be traded against each other as shown by previous studies.25 Plots of each of these curves created a Pareto front. Better plan sets were defined as those with a front closest to the axes. This analysis was used to show any systematic differences in plan quality following conversion of navigated plans to deliverable plans.
Planning analysis Standard inverse plan generation An initial IMRT plan was designed to deliver a uniform dose to all 3 target structures at their respective prescription level, while imposing no objective or constraint on sparing of the rectum. Objectives that penalized deviations from a uniform dose at 100%, 96%, and 80% of prescription were assigned to PTV 3, PTV 2 PTV 3, and PTV 1 PTV 2, respectively. Objectives were introduced to reduce the dose to the avoidance shell, which reduced dose to the bladder and increased conformity. Objectives were also introduced that required PTV 3 to satisfy D100% Z 54 Gy and D95% Z 57 Gy, PTV 2 PTV 3 to satisfy D100% Z 52 Gy and D5% r 61 Gy, and PTV 2 PTV 1 to satisfy D100% Z 46 Gy and D5% r 57 Gy. Reoptimizations were subsequently performed also in the presence of an objective on the average dose to the rectum. The weight of this objective was increased in incremental steps until the minimum average rectal dose was achieved with an acceptable plan following the tolerance levels specified by the CHHiP protocol (Table 1) and manual inspection of DVHs planned dose distributions.
The clinically acceptable plan with minimal average rectal dose obtained for each patient case and planning technique were analyzed. Clinical acceptability was ascertained by both the constraints as indicated by the CHHiP trial tolerances and the dose coverage as determined by a radiation oncologist. RaySearch MCO plans that were acceptable at the navigated stage and at the deliverable stage (following interpretation) were termed navigated deliverable. These were compared with the Oncentra standard IMRT plans in the first instance. Similarly, the deliverable MCO plan satisfying CHHiP requirements and having minimum rectum dose was compared with a standard IMRT plan to assess the effect of eliminating errors introduced by the navigation to delivery step (best deliverable). The evaluation was performed by a pair-wise comparison of median and first and third quartiles over a set of performance metrics on the spatial dose distribution. Statistical analysis was undertaken comparing the conventionally optimized plans and the MCO plans using Wilcoxon signed rank test with a level of significance placed at p ¼ 0.05 (SPSS 15.0.1.1). The same analysis was used to compare the navigated and deliverable MCO plans.
MCO planning algorithm and plan generation Databases of fluence-based MCO plans were generated for each patient case with respect to a set of preselected trade-off objectives and hard constraints. Objectives that penalized deviations from a uniform dose at 100%, 96%, and 80% of prescriptions were assigned to PTV 3, PTV 2 PTV 3, and PTV 1 PTV 2, respectively. Objectives were also introduced that strived toward reducing the dose to the rectum and avoidance shell. Both these objectives were posed as a direct penalty on average dose. As with standard optimization, dose to the bladder was only accounted for indirectly through control of the dose to the avoidance shell. Hard constraints were introduced that required PTV 3 to satisfy D100% Z 54 Gy and
Results Planning and delivery analysis The dose distributions calculated by both PB (Oncentra) and CC (RayStation) had 100% pixels passing the gamma 3% 3-mm criteria following delivery of a clinically acceptable DSS optimized treatment plan (patient 2) to the MatriXX 2D ionization array. For the
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]]
4
Table 2 Comparison of dosimetric parameters between 5-field IMRT DSS plans and MCO plans for 10 patients with prostate cancer Parameters
Standard IMRT
MCO (navigated deliverable)
Wilcoxon signed rank test (p value)
MCO (best deliverable)
Wilcoxon signed rank test (p value)
HI PTV 1 HI PTV 2 HI PTV 3 CI PTV 1 CI PTV 2 CI PTV 3 Ave rectal dose (Gy) R41 (%) R54 (%) R68 (%) R81 (%) R88 (%) R95 (%) R100 (%) B68 (%) B81 (%) B100 (%) P_B68 (%) P_B81 (%)
0.199 0.076 0.056 1.686 1.251 1.752 35.5 84.7 62.3 33.1 18.6 11.6 5.50 0.10 15.6 11.2 0.88 57.0 27.5
0.165 0.051 0.027 1.731 1.336 1.901 33.7 78.1 52.1 32.7 18.2 11.5 5.29 0.00 17.8 12.9 0.28 51.7 37.7
0.007 0.005 0.005 0.508 0.007 0.007 0.114 0.203 0.074 0.646 0.333 0.074 0.799 0.018 0.022 0.005 0.007 0.139 0.310
0.164 0.055 0.032 1.809 1.361 1.893 30.6 66.9 40.5 29.5 17.6 11.3 5.19 0.01 17.8 12.9 0.52 55.7 40.4
0.028 0.008 0.007 0.074 0.007 0.013 0.047 0.017 0.037 0.139 0.169 0.028 0.959 0.025 0.005 0.005 0.047 0.508 0.063
(0.180 to 0.206) (0.069 to 0.083) (0.043 to 0.057) (1.558 to 1.839) (1.195 to 1.295) (1.623 to 1.841) (31.3 to 39.2) (76.7 to 89.9) (39.9 to 73.7) (19.0 to 47.3) (10.8 to 25.2) (7.8 to 17.6) (2.64 to 7.52) (0.00 to 0.22) (11.2 to 25.9) (6.5 to 16.9) (0.16 to 1.78) (8.3 to 85.0) (0.0 to 57.2)
(0.155 to 0.192) (0.047 to 0.054) (0.023 to 0.029) (1.669 to 1.801) (1.317 to 1.399) (1.791 to 1.935) (31.9 to 36.1) (74.1 to 81.6) (38.35 to 60.0) (22.4 to 39.9) (11.5 to 22.9) (7.0 to 14.7) (2.81 to 7.76) (0.00 to 0.00) (12.5 to 28.4) (8.6 to 20.0) (0.01 to 0.68) (10.8 to 67.8) (0.0 to 52.3)
(0.154 to 0.188) (0.047 to 0.063) (0.023 to 0.040) (1.741 to 1.891) (1.330 to 1.583) (1.803 to 1.963) (29.2 to 36.1) (59.8 to 81.6) (31.2 to 60.0) (19.4 to 39.0) (11.4 to 22.6) (6.8 to 14.25) (2.73 to 7.08) (0.00 to 0.03) (13.7 to 28.7) (9.9 to 21.1) (0.07 to 1.33) (10.8 to 70.0) (0.00 to 56.0)
Both p values are comparisons between MCO and standard IMRT. Shown are the median, first, and third quartiles. Wilcoxon signed rank tests performed for each with significance level at p ¼ 0.05. Significant p-values are indicated in bold. CI ¼ conformity index; R ¼ rectum; B ¼ bladder; P_B ¼ penile bulb. Subscripts show percentage dose point that volume on DVH is analyzed.
Oncentra PB algorithm and RayStation CC algorithm, the average gammas were 0.24 ⫾ 0.15 and 0.22 ⫾ 0.12, respectively. Plots of average rectal dose against HI for both the calculation algorithms for all the patients showed minimal differences across the entire Pareto front (data not shown). Table 2 summarizes dosimetric data attained from the DVH of the conventional IMRT plan and the MCO plans that satisfied CHHiP requirements and had minimum rectum dose for each patient. The tabulated data show that the target volumes of the MCO plans were more homogeneous than those of the conventionally optimized plans for all 3 PTVs. The average rectal dose and rectal dose points were also lower for the MCO plans. The pattern was reversed if comparing the planned dose to the bladder and dose conformity to the target volumes, where better results were observed for the conventionally optimized plans. Conformity was not statistically different for dose surrounding all the PTVs (as depicted by PTV 1), although differences were observed for the boost regions. The clinically acceptable MCO plans had lower rectal doses for almost all tolerance levels and for the dose to 100% of the bladder volume. The dose received by 81% and 68% of the bladder volume was significantly less for the conventionally optimized plans. The best possible clinically acceptable MCO plans, irrespective of the navigated plan, followed a similar trend, although the average dose to the rectum reduced further. Pareto front analysis Figure 2 shows Pareto fronts of target homogeneity vs average rectal dose for all navigated plans (fluence space) of each patient plotted along with the same plans, after they had been converted to deliverable MLC apertures and accurate dose calculated with CC. In the region of the Pareto front where the average rectal dose was largest, the difference between the navigated and deliverable plan was minimal although the difference progressively increased as the average rectal dose reduced. The navigated Pareto front is closest to the axis, which would suggest that this plan set is of superior quality compared with the deliverable plan. Figure 3 shows a representative example of a DVH distribution for a clinically acceptable MCO plan (patient 10) both in the navigated state (solid line) and after a conversion into deliverable MLC segments with the final dose calculated with CC (dashed line).
Differences between the DVH plot of PTV 3 are minor overall, but nevertheless relevant because they occur close to the dose received by 99% of the target volume, which can have a significant effect on target coverage when viewing dose distributions. The differences between the navigated and deliverable DVH plots of the rectum are of larger magnitude. RayStation permits the user to select a scalarvalued target priority weight that determines how replication of the navigated DVH in target structures is traded against replicating DVH curves corresponding to healthy tissue. The dashed lines indicate deliverable DVHs obtained with a target priority of 30, which is the system default and the setting used for all numerical results presented in this study. The sensitivity with respect to this parameter is illustrated by the dotted DVHs that correspond to a deliverable plan generated with a target priority of 50. As expected, a higher priority to targets results in improved target coverage but at the expense of large deviations for the critical structures. A final assessment was performed regarding to what extent clinical decisions can be made on the basis of navigated dose distributions alone. This analysis showed that the navigated plans satisfying CHHiP requirements and having minimum rectum dose rarely translate to the acceptable deliverable plans following a final dose calculation using RayStation or Oncentra. The acceptable navigated plan with minimum rectum dose could be converted to an acceptable deliverable plan using RayStation CC algorithm for 7 of 10 patients. Also, the acceptable navigated plan with minimal rectum dose could be converted to an acceptable deliverable plan on Oncentra PB algorithm for 5 of 10 patients. On a number of occasions navigated plans were not clinically acceptable owing to either lack of target coverage or OAR tolerances not being met but were, in fact, acceptable following conversion to a deliverable plan. The best deliverable column in Table 2 shows plans satisfying CHHiP requirements and having minimum rectum dose irrespective of the acceptability of the initial plan navigation. There was a clinically acceptable plan with a lower rectal dose in 6 of 10 patients if the plan exactly matched the navigated plans compared with plans that required a clinically acceptable navigated plan. Discussion In this study, we have validated RayStation MCO against Oncentra standard IMRT optimization. MCO and standard IMRT
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]]
Fig. 2. Pareto fronts of 10 patients comparing multicriteria optimization (MCO) navigation (MCO Navigated, plan following segmentation and collapsed-cone final dose calculation (MCO deliverable, , solid line).
▲
5
△, long dashed line) with the corresponding deliverable MCO
6
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]]
Fig. 3. DVH of patient 10 with MCO navigated plan (solid) and the final deliverable plan after collapsed-cone final dose calculation optimized with target priority 30 (dashed) and 50 (dotted). (Color version of figure is available online.)
plans were observed to be of similar quality. The MCO plans also showed similar or superior target homogeneity compared with the conventionally optimized IMRT plans. Rectal sparing of the MCO plans were to some extent at a cost of less conformal dose distributions and higher bladder doses. The analysis of the dose deviations between navigated and final deliverable MCO plans revealed that sacrifices in dose quality owing to the conversion was primarily made in terms of sparing of critical structures. In addition, these dose differences were found to increase as a higher priority was placed on avoidance of the rectum. Differences between the final dose calculation methods of the 2 independent planning systems used in this study (i.e., PB in Oncentra and CC in RayStation) only resulted in very minor perturbations of the Pareto fronts, and these differences were also confirmed to be minor in experimental measurements of the accuracy of delivery. Slight improvements in plan quality would be demonstrated if the navigation to delivery step was eliminated. Inverse planning for IMRT delivery is well known to often require a significant amount of manual parameter tuning before a final clinically acceptable plan is settled upon by the planner. This plan may then be checked by the clinician who may then ask for further refinement depending on the quality of the plan presented. MCO can significantly improve the efficiency of this process as the database of possible plans can be presented to and navigated by the clinician before settling on a specific plan.11,26 It is clear from this study that MCO plans of similar quality to standard IMRT plans can be created for patients with prostate cancer where stringent tolerances are often used. The flexibility achieved by having the option of navigating to the required plan by a clinician can also undoubtedly improve planning efficiency. On the negative side, our results show that the dose deviations that occur because of the conversion of a navigated treatment plan into deliverable machine setting in some cases can render an initially acceptable treatment plan unacceptable for clinical delivery. Such dose deviations would in a clinical setting need to be counteracted by parameter adjustments followed by generation of another deliverable treatment plan; a process that defeats the initial purpose of MCO. In this study, the main dose deviation because of the conversion process resulted in either lack of target coverage or hotspots in the targets. Significant increases were also observed in the dose to critical structures. The Pareto fronts analysis in Fig. 2 clearly indicates that there is potential for the navigated plans to better approximate the
deliverable frontier as there appears to be a systematic shift of the deliverable curve from the navigated curve, whereby the differences get larger at lower average rectal doses. This result was consistent in all the 10 patients. Further analysis also showed that the navigated plan satisfying CHHiP requirements and having minimum rectum dose may not result in an acceptable deliverable plan. For a number of patients more acceptable MCO plans could have been utilized but because the navigated plans were unacceptable they would have been discounted by the clinician. Better characterization of the navigated plans would constitute a solution to this problem. Prostate plans are not the most complex IMRT plans to create,19 although placing a high priority on avoidance of the rectum does increase the complexity of the plan. It appears from the data presented in this study that the more complex initial navigated plans (with lower average rectal dose) deviate most from the final deliverable plans. Hence, it may be important to validate any new approximations, developed for use with MCO planning, using both the multidose level IMRT treatment on a cohort of patients with prostate cancer presented in this article, as well as more complex cases, such as patients with head and neck cancer. Further work is underway through improved algorithms to generate base or navigated plans that correlate more closely with deliverable plans. Examples of such efforts include navigation among segmented plans where each plan has unique MLC shapes and restrictions are enforced regarding how plans can be combined,13,14,27 and unrestricted navigation among segmented base plans that have shared MLC shapes.22 It is important that these are clinically validated, and these should then enhance the current power of MCO planning.
Conclusions MCO appears to have the potential to create equivalent or better plans for patients with prostate cancer, than those that are created using standard inverse planning methods. Further improvements in MCO planning may be possible through better approximation of deliverable plans during the navigation phase of MCO planning.
Acknowledgments The authors thank Björn Hårdemark, RaySearch Laboratories, for his helpful comments on the article. References 1. Sheets, N.C.; Goldin, G.H.; Meyer, A.M.; et al. Intensity-modulated radiation therapy, proton therapy, or conformal radiation therapy and morbidity and disease control in localized prostate cancer. J. Am. Med. Assoc. 307(15):1611–20; 2012. 2. Bohsung, J.; Gillis, S.; Arrans, R.; et al. IMRT treatment planning:- a comparative inter-system and inter-centre planning exercise of the ESTRO QUASIMODO group. Radiother. Oncol. 76(3):354–61; 2005. 3. Chung, H.T.; Lee, B.; Park, E.; et al. Can all centers plan intensity-modulated radiotherapy (IMRT) effectively? An external audit of dosimetric comparisons between three-dimensional conformal radiotherapy and IMRT for adjuvant chemoradiation for gastric cancer Int. J. Radiat. Oncol. Biol. Phys. 71(4):1167–74; 2008. 4. Seco, J.; Clark, C.H.; Evans, P.M.; et al. A quantitative study of IMRT delivery effects in commercial planning systems for the case of oesophagus and prostate tumours. Br. J. Radiol. 79(941):401–8; 2006. 5. Shepard, D.M.; Earl, M.A.; Li, X.A.; et al. Direct aperture optimization: a turnkey solution for step-and-shoot IMRT. Med. Phys. 29(6):1007–18; 2002. 6. Craft, D.; Halabi, T.; Bortfeld, T. Exploration of tradeoffs in intensity-modulated radiotherapy. Phys. Med. Biol. 50(24):5857–68; 2005. 7. Hoffman, A.; Siem, A.; den Hertog, D.; et al. Derivative-free generation and interpolation of convex Pareto optimal IMRT plans. Phys. Med. Biol. 51 (24):6349–69; 2006. 8. Craft, D.; Halabi, T.; Shih, H.A.; et al. An approach for practical multiobjective IMRT treatment planning. Int. J. Radiat. Oncol. Biol. Phys. 69(5):1600–7; 2007.
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]] 9. Serna, J.; Monz, M.; Küfer, K.-H.; et al. Trade-off bounds for the Pareto surface approximation in multi-criteria IMRT planning. Phys. Med. Biol. 54(20): 6299–311; 2009. 10. Hong, T.S.; Craft, D.L.; Carlsson, F.; et al. Multicriteria optimization in intensitymodulated radiation therapy treatment planning for locally advanced cancer of the pancreatic head. Int. J. Radiat. Oncol. Biol. Phys. 72(4):1208–14; 2008. 11. Craft, D.L.; Hong, T.S.; Shih, H.A.; et al. Improved planning time and plan quality through multicriteria optimization for intensity-modulated radiotherapy. Int. J. Radiat. Oncol. Biol. Phys. 82(1):e83–90; 2012. 12. Dearnaley, D.; Syndikus, I.; Sumo, G.; et al. Conventional versus hypofractionated high-dose intensity-modulated radiotherapy for prostate cancer: preliminary safety results from the CHHiP randomised controlled trial. Lancet Oncol. 13:43–54; 2012. 13. Salari, E.; Unkelbach, J. A column-generation-based method for multi-criteria direct aperture optimization. Phys. Med. Biol. 58(3):621–39; 2013. 14. Craft, D.; Richter, C. Deliverable navigation for multicriteria step and shoot IMRT treatment planning. Phys. Med. Biol. 58(1):87–103; 2013. 15. Philips, M.; Holdsworth, C. When is better best? A multiobjective perspective Med. Phys. 38(3):1635–40; 2011. 16. Ottosson, R.O.; Engstrom, P.E.; Sjöström, D.; et al. The feasibility of using Pareto fronts for comparison of treatment planning systems and delivery techniques. Acta Oncol. 48(2):233–7; 2009. 17. Petersson, K.; Ceberg, C.; Engström, P.; et al. Beam commissioning and measurements validating the beam model in a new TPS that converts helical tomotherapy plans to step-and-shoot IMRT plans. Med. Phys. 38(1):40–6; 2011. 18. Pardo-Montero, J.; Fenwick, J. Tomotherapy-like versus VMAT-like treatments: a multicriteria comparison for a prostate geometry. Med. Phys. 39(12):7418–29; 2012.
7
19. McGarry, C.K.; Chinneck, C.D.; O’Toole, M.M.; et al. Assessing software upgrades, plan properties and patient geometry using intensity modulated radiation therapy (IMRT) complexity metrics. Med. Phys. 38(4):2027–34; 2011. 20. Bokrantz, R.; Forsgren, A. An algorithm for approximating convex Pareto surfaces based on dual techniques. INFORMS J. Comput. 25(2):377–93; 2013. 21. Hårdemark, B.; Liander, A.; Rehbinder, H.; et al. Direct machine parameter optimization with RayMachines in Pinnacle3s. White paper, RaySearch Laboratories, Stockholm, Sweden; 2003. 22. Bokrantz, R. Multicriteria optimization for volumetric-modulated arc therapy by decomposition into a fluence-based relaxation and a segment weight-based restriction. Med. Phys. 39(11):6712–25; 2012. 23. Bortfeld, T.; Schlegel, W.; Rhein, B. Decomposition of pencil beam kernels for fast dose calculations in three‐dimensional treatment planning. Med. Phys. 20 (2):311–8; 1993. 24. Mott, J.H.; Livsey, J.E.; Logue, J.P. Development of a simultaneous boost IMRT class solution for a hypofractionated prostate cancer protocol. Br. J. Radiol. 77 (917):377–86; 2004. 25. McGarry, C.K.; McMahon, S.J.; Craft, D.; et al. Inverse planned constant dose rate volumetric modulated arc therapy (VMAT) as an efficient alternative to fivefield intensity modulated radiation therapy (IMRT) for prostate. J. Radiother. Pract. 13(1):68–78; 2014. 26. Wala, J.; Craft, D.; Paly, J.; et al. Maximizing dosimetric benefits of IMRT in the treatment of localized prostate cancer through multicriteria optimization planning. Med. Dosim. 38(3):298–303; 2013. 27. Fredriksson, A.; Bokrantz, R. Deliverable navigation for multicriteria IMRT treatment planning by combining shared and individual apertures. Phys. Med. Biol. 58(21):7683–97; 2013.
Medical Dosimetry journal homepage: www.meddos.org
Advantages and limitations of navigation-based multicriteria optimization (MCO) for localized prostate cancer IMRT planning Conor K. McGarry, Ph.D.,* Rasmus Bokrantz, Ph.D.,†‡ Joe M. O’Sullivan, F.F.R. (RCSI),§║ and Alan R. Hounsell, Ph.D.*§ Radiotherapy Physics, Northern Ireland Cancer Centre, Belfast Health and Social Care Trust, Belfast, Northern Ireland, UK; †Optimization and Systems Theory, KTH Royal Institute of Technology, Stockholm, Sweden; ‡RaySearch Laboratories, Stockholm, Sweden; §Centre for Cancer Research and Cell Biology, Queen’s University Belfast, Belfast, Northern Ireland, UK; and ║Clinical Oncology, Northern Ireland Cancer Centre, Belfast Health and Social Care Trust, Belfast, Northern Ireland, UK
n
A R T I C L E I N F O
A B S T R A C T
Article history: Received 22 October 2013 Accepted 3 February 2014
Efficacy of inverse planning is becoming increasingly important for advanced radiotherapy techniques. This study’s aims were to validate multicriteria optimization (MCO) in RayStation (v2.4, RaySearch Laboratories, Sweden) against standard intensity-modulated radiation therapy (IMRT) optimization in Oncentra (v4.1, Nucletron BV, the Netherlands) and characterize dose differences due to conversion of navigated MCO plans into deliverable multileaf collimator apertures. Step-and-shoot IMRT plans were created for 10 patients with localized prostate cancer using both standard optimization and MCO. Acceptable standard IMRT plans with minimal average rectal dose were chosen for comparison with deliverable MCO plans. The trade-off was, for the MCO plans, managed through a user interface that permits continuous navigation between fluence-based plans. Navigated MCO plans were made deliverable at incremental steps along a trajectory between maximal target homogeneity and maximal rectal sparing. Dosimetric differences between navigated and deliverable MCO plans were also quantified. MCO plans, chosen as acceptable under navigated and deliverable conditions resulted in similar rectal sparing compared with standard optimization (33.7 ⫾ 1.8 Gy vs 35.5 ⫾ 4.2 Gy, p ¼ 0.117). The dose differences between navigated and deliverable MCO plans increased as higher priority was placed on rectal avoidance. If the best possible deliverable MCO was chosen, a significant reduction in rectal dose was observed in comparison with standard optimization (30.6 ⫾ 1.4 Gy vs 35.5 ⫾ 4.2 Gy, p ¼ 0.047). Improvements were, however, to some extent, at the expense of less conformal dose distributions, which resulted in significantly higher doses to the bladder for 2 of the 3 tolerance levels. In conclusion, similar IMRT plans can be created for patients with prostate cancer using MCO compared with standard optimization. Limitations exist within MCO regarding conversion of navigated plans to deliverable apertures, particularly for plans that emphasize avoidance of critical structures. Minimizing these differences would result in better quality treatments for patients with prostate cancer who were treated with radiotherapy using MCO plans. & 2014 American Association of Medical Dosimetrists.
Keywords: IMRT MCO Prostate
Background Intensity-modulated radiation therapy (IMRT) has been shown to be superior to conformal radiation therapy for patients with prostate cancer.1 Generating high-quality IMRT plans is, however,
Reprint requests to: Conor Kevin McGarry, Ph.D., Radiotherapy Physics, Northern Ireland Cancer Centre, Belfast Health and Social Care Trust, Lisburn Road, Belfast BT9 7AB, UK. Tel.: þ44 2895 048 327. E-mail: [email protected] http://dx.doi.org/10.1016/j.meddos.2014.02.002 0958-3947/ Copyright Ó 2014 American Association of Medical Dosimetrists.
often a time-consuming process because the priorities of conflicting planning goals need to be adjusted and the treatment plan reoptimized multiple times, until it is acceptable for clinical delivery. The quality of plans generated by standard methods has also been shown to depend on both experience level and time commitment of the individual treatment planner.2,3 A further difficulty is that differences between optimized and deliverable treatment plans can occur, especially when the plan requires interpretation of multileaf collimator (MLC) leaf positions.4 Developments such as direct-aperture optimization have minimized
2
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]]
these differences by including the MLC positions and associated limitations within the optimization step of creating an IMRT plan.5 Improvements in treatment planning efficacy for advanced radiotherapy techniques are becoming more important with the onset of adaptive radiotherapy. Multicriteria optimization (MCO) represents a significant step toward minimizing the number of manual iterations or recalculations during the creation of an IMRT plan.6-9 This technology utilizes a database of precomputed IMRT plans whose dose distributions are continuously combined to produce a single “navigated” treatment plan. Dose averaging over the database plans is performed according to the input from a graphical user interface that contains a set of slider bar controls that each represents the priority of a single planning objective.8,9 Navigation is feasible in real time and, thereby, permits the treatment planner to obtain immediate feedback regarding how a change of the priority of one objective correlates with the other objectives as well as how it affects the 3-dimensional dose distribution. MCO of this form has, for a number treatment sites, been shown to significantly reduce treatment planning times and simultaneously result in treatment plans that physicians judged as superior to conventionally optimized IMRT plans in blinded assessments.10,11 However, there is a paucity of data on comparisons between MCO and standard planning in terms of quantitative statistics on planned physical dose distributions for patients with prostate cancer. To complement the findings by previous authors that MCO enables planners to more easily find a suitable trade-off between objectives, this study, therefore, sought to quantify the costs and benefits of MCO with respect to dose distribution quality. This investigation is performed with respect to a suite of patients with localized prostate cancer with multiple target regions at different dose levels, as prescribed by the conventional or hypofractionated high dose IMRT trial for prostate cancer (CHHiP) protocol.12 As a secondary goal, this study also addressed the fact that a predominant number of approaches to MCO in the literature rely on databases of fluence-based treatment plans.6-9 This property can be traced to the fact that the set of database plans that form the basis for the navigated dose in MCO are computed toward forming a representation of the set of Pareto optimal treatment plans, i.e., feasible treatment plans that are nondominated in the sense that there does not exist another feasible plan that is strictly better in at least one objective and no worse in every other objective. A fluence-based representation of this set (that is subsequently called the Pareto front for short) has the advantage that there exists a one-to-one correspondence between a linear combination in dose and the corresponding linear combination in fluence. By virtue of this fact, a combination of multiple Pareto optimal fluence profiles is guaranteed to produce another feasible and near-optimal fluence-based treatment plan. This property does not extend to databases that contain deliverable treatment plans because there is no trivial way of combing multiple MLC segments into a single treatment plan with dose identical to the corresponding combination of dose distributions without increasing the total number of segments. Current research is attempting to address this issue through multicriteria direct-aperture optimization13 or segmentation during the navigation phase.14 A fluencebased representation, however, necessitates a conversion into deliverable MLC segments that can compromise plan quality. We critically assess the dose discrepancies that occur owing to this conversion and study how the magnitude of the conversion errors correlates with the spatial shape of the navigated dose distributions. Pareto fronts are not only useful as a model of the relevant treatment options, but can also be used as a form of assessment of plan quality that is not biased by subjective decisions on which
aspects of plan quality to prioritize.15 Such comparisons have previously been applied to the assessment of the effect of different planning parameters for a fixed delivery technique16 as well as assessment of differences between alternate treatment modalities.17,18 In this study, we assessed plan quality of the MCO plans for patients with prostate cancer compared with the standard IMRT plan. The accuracy of MCO plan delivery was validated. We also examined plan quality of treatment plans generated by navigated vs the corresponding deliverable MCO plans by comparison of the Pareto front spanned by target homogeneity vs average dose to the rectum.
Methods and Materials Patients Plans were created using computed tomography (CT) datasets of 10 successive patients with prostate cancer (Ethics REC ref 09/NIR02/28). Each patient had been CT scanned using a GE LightSpeed wide-bore scanner (GE Healthcare, WI) at 2.5mm slices. Target volumes were expanded and organs at risk (OARs) delineated, including the rectum, bladder, femoral heads, penile bulb, and bowel, according to the CHHiP protocol.12 Plans were created with a prescription of 60 Gy in 20 fractions, a schedule that has been used in previous planning studies.25 Dose levels to the targets (planning target volumes [PTVs] 1 through 3) and tolerances for OARs are shown in Table 1. An avoidance shell (1-cm thick) was delineated, which ensured a steep fall-off of dose from the targets to the surrounding area, as illustrated in Fig. 1. The avoidance shell was designed to not involve the rectum to ensure that target homogeneity was traded against sparing of the rectum.
Treatment planning Standard IMRT treatment plans were optimized on a single PC (HP z8000 Workstation, 2 Intel Xeon CPU X5660 2.8 GHz and 12 GB of RAM) using the direct step-and-shoot (DSS) optimization algorithm of the Oncentra treatment planning system v4.1 (Nucletron BV, Veenendal, the Netherlands). An enhanced pencil-beam (PB) dose calculation algorithm was used for final dose calculations. This dose algorithm includes modeling of the MLC type, MLC transmission, and source size. The PB algorithm was used as it has been fully validated in our department as well as other departments.19 Treatment plans were designed for delivery on a Varian 2100CD linear accelerator equipped with 120-leaf millennium MLC (Varian Medical Systems, Palo Alto, CA) via the ARIA record and verify system, using 6-MV photons. MCO plans were created on a single PC (HP EliteBook 8530w, 2 Intel Core CPU T9400 2.53 GHz and 6 GB of RAM) using the RayStation treatment planning system v2.4 (RaySearch Laboratories, Stockholm, Sweden). The database of fluence-based treatment plans that form the basis of a navigated MCO plan is in this system calculated toward minimizing the approximation error of the database representation with respect to the underlying Pareto front.20 Navigation is performed by realtime linear interpolation of dose over the database plans. A navigated treatment plan selected for conversion by the user is converted to deliverable MLC segments through DSS optimization.21 The objective of this optimization is to minimize the error between the dose-volume histogram (DVH) distributions of the navigated and the deliverable treatment plan.22 Dose for the MCO plans was computed by an algorithm that implements singular value decomposition (SVD) of PB kernels.23 An intermediate and a final dose calculation were in addition performed during DSS optimization using a collapsed-cone (CC) dose algorithm.21 The intermediate CC dose was used as background dose for subsequent SVD dose calculations. Dose calculations using SVD and CC were both performed with respect to a machine Table 1 Selection of the dosimetric quality parameters outlined in the CHHiP trial protocol CHHiP trial parameter
Constraint (%)
CHHiP trial parameter
Constraint
PTV 1 min PTV 2 min PTV 3 min Rectum v68% max Rectum v81% max Rectum v88% max Rectum v95% max Rectum v100% max
76 91 95 60 50 30 15 3
Bladder v68% max Bladder v81% max Bladder v100% max Penile bulb v68% max Penile bulb v81% max Bowel v68% max Femoral heads v68% max
50% 25% 5% 50% 10% 17 cc 50%
PTV min is the minimum dose (%) received by 99% of the volume. For critical structures, the constraint is the maximum dose received by the volume X where v(X)max is shown (e.g., X ¼ 68 for femoral heads).
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]]
3
Fig. 1. Targets outlined to receive at least 76% (PTV 1), 91% (PTV 2), and 95% (PTV 3) of prescribed dose with avoidance of the rectum shown. A shell (red), not including the rectum was included to ensure conformity. (Color version of figure is available online.) model commissioned according to experimental beam data for the linear accelerator used for this study. Standard IMRT plans, as well as MCO plans, were planned toward step-andshoot delivery with a dose rate of 400 MU/min. A 5-field class solution was used with beam angles of 351, 1001, 1801, 2601, and 3251 for both planning methods. This class solution has been developed from beam angles published previously.24 The number of segments was 50 per plan, the minimum field size was 4 cm2, and the grid size was set to 0.25 cm in all directions. All plans were assessed in terms of acceptability according to the CHHiP protocol.12 Each of the indices specified in Table 1 were also compared across all the plans. Further analysis was performed on targets in the form of homogeneity index (HI) to PTV 3 defined as D5% D95% (where D5% is the dose to 5% volume and D95% is dose to 95% volume), and conformity index calculated as the volume of the body receiving the prescribed dose divided by the corresponding PTV. The lower the conformity index, the more conformal the plan is expected to be, and the lower the HI the more uniform the dose delivered to the target.
Dose calculation validation As 2 different dose calculation engines were used between the MCO plans and the standard plans generated using Oncentra, one of the plans was tested for delivery accuracy. A deliverable plan (patient 2—generated by DSS optimization in Oncentra) was recalculated on a CT scan set of a MatriXX Evolution 2D ionization chamber array device (IBA, Schwarzenbruck, Germany), with a 6-cm thick 30 30cm2 solid water slab above and below the device, using the CC algorithm in RayStation and the PB algorithm in Oncentra. Both plans were delivered to the array using a Varian 2100CD linear accelerator, and the measured data in the coronal plane were compared with that calculated using gamma analysis with the dose and distance criteria set at 3% and 3 mm, respectively.
D95% Z 57 Gy, PTV 2 PTV 3 to satisfy D100% Z 52 Gy and D5% r 61 Gy, and PTV 2 PTV 1 to satisfy D100% Z 46 Gy and D5% r 57 Gy. A dose-limiting constraint on the global maximum dose was finally introduced at 63 Gy. The stated constraints collectively ensure that the database representation of the Pareto front is restricted to clinically relevant treatment plans. A Pareto front of target homogeneity vs average rectal dose was, after database generation, extracted by navigation over the database plans. A navigated plan that gave maximum priority to target homogeneity and that had priorities for the other goals set to ensure plan acceptability was first created. The priority to rectal sparing was subsequently increased in incremental steps (through a movement of the slider bar control associated with this objective), while adjustments were performed using the slider bars of the other objectives to retain acceptability of the created plan when possible. A series of navigation states at discrete steps along the trajectory between maximal homogeneity and maximal rectal sparing were saved as unique treatment plans. These plans were subsequently assessed for acceptability analogous to the assessment of the conventionally optimized plans. Each navigated plan was converted into deliverable MLC segments and final dose computed by CC. To facilitate a fair comparison with the conventionally optimized plans, the deliverable plans were transferred to Oncentra where the final dose was recalculated using the PB dose algorithm, and the plans again assessed for acceptability. To analyze changes between navigated and deliverable plans, plots were created of average rectal dose against homogeneity of PTV, defined using the HI. These were considered to be 2 factors that could be traded against each other as shown by previous studies.25 Plots of each of these curves created a Pareto front. Better plan sets were defined as those with a front closest to the axes. This analysis was used to show any systematic differences in plan quality following conversion of navigated plans to deliverable plans.
Planning analysis Standard inverse plan generation An initial IMRT plan was designed to deliver a uniform dose to all 3 target structures at their respective prescription level, while imposing no objective or constraint on sparing of the rectum. Objectives that penalized deviations from a uniform dose at 100%, 96%, and 80% of prescription were assigned to PTV 3, PTV 2 PTV 3, and PTV 1 PTV 2, respectively. Objectives were introduced to reduce the dose to the avoidance shell, which reduced dose to the bladder and increased conformity. Objectives were also introduced that required PTV 3 to satisfy D100% Z 54 Gy and D95% Z 57 Gy, PTV 2 PTV 3 to satisfy D100% Z 52 Gy and D5% r 61 Gy, and PTV 2 PTV 1 to satisfy D100% Z 46 Gy and D5% r 57 Gy. Reoptimizations were subsequently performed also in the presence of an objective on the average dose to the rectum. The weight of this objective was increased in incremental steps until the minimum average rectal dose was achieved with an acceptable plan following the tolerance levels specified by the CHHiP protocol (Table 1) and manual inspection of DVHs planned dose distributions.
The clinically acceptable plan with minimal average rectal dose obtained for each patient case and planning technique were analyzed. Clinical acceptability was ascertained by both the constraints as indicated by the CHHiP trial tolerances and the dose coverage as determined by a radiation oncologist. RaySearch MCO plans that were acceptable at the navigated stage and at the deliverable stage (following interpretation) were termed navigated deliverable. These were compared with the Oncentra standard IMRT plans in the first instance. Similarly, the deliverable MCO plan satisfying CHHiP requirements and having minimum rectum dose was compared with a standard IMRT plan to assess the effect of eliminating errors introduced by the navigation to delivery step (best deliverable). The evaluation was performed by a pair-wise comparison of median and first and third quartiles over a set of performance metrics on the spatial dose distribution. Statistical analysis was undertaken comparing the conventionally optimized plans and the MCO plans using Wilcoxon signed rank test with a level of significance placed at p ¼ 0.05 (SPSS 15.0.1.1). The same analysis was used to compare the navigated and deliverable MCO plans.
MCO planning algorithm and plan generation Databases of fluence-based MCO plans were generated for each patient case with respect to a set of preselected trade-off objectives and hard constraints. Objectives that penalized deviations from a uniform dose at 100%, 96%, and 80% of prescriptions were assigned to PTV 3, PTV 2 PTV 3, and PTV 1 PTV 2, respectively. Objectives were also introduced that strived toward reducing the dose to the rectum and avoidance shell. Both these objectives were posed as a direct penalty on average dose. As with standard optimization, dose to the bladder was only accounted for indirectly through control of the dose to the avoidance shell. Hard constraints were introduced that required PTV 3 to satisfy D100% Z 54 Gy and
Results Planning and delivery analysis The dose distributions calculated by both PB (Oncentra) and CC (RayStation) had 100% pixels passing the gamma 3% 3-mm criteria following delivery of a clinically acceptable DSS optimized treatment plan (patient 2) to the MatriXX 2D ionization array. For the
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]]
4
Table 2 Comparison of dosimetric parameters between 5-field IMRT DSS plans and MCO plans for 10 patients with prostate cancer Parameters
Standard IMRT
MCO (navigated deliverable)
Wilcoxon signed rank test (p value)
MCO (best deliverable)
Wilcoxon signed rank test (p value)
HI PTV 1 HI PTV 2 HI PTV 3 CI PTV 1 CI PTV 2 CI PTV 3 Ave rectal dose (Gy) R41 (%) R54 (%) R68 (%) R81 (%) R88 (%) R95 (%) R100 (%) B68 (%) B81 (%) B100 (%) P_B68 (%) P_B81 (%)
0.199 0.076 0.056 1.686 1.251 1.752 35.5 84.7 62.3 33.1 18.6 11.6 5.50 0.10 15.6 11.2 0.88 57.0 27.5
0.165 0.051 0.027 1.731 1.336 1.901 33.7 78.1 52.1 32.7 18.2 11.5 5.29 0.00 17.8 12.9 0.28 51.7 37.7
0.007 0.005 0.005 0.508 0.007 0.007 0.114 0.203 0.074 0.646 0.333 0.074 0.799 0.018 0.022 0.005 0.007 0.139 0.310
0.164 0.055 0.032 1.809 1.361 1.893 30.6 66.9 40.5 29.5 17.6 11.3 5.19 0.01 17.8 12.9 0.52 55.7 40.4
0.028 0.008 0.007 0.074 0.007 0.013 0.047 0.017 0.037 0.139 0.169 0.028 0.959 0.025 0.005 0.005 0.047 0.508 0.063
(0.180 to 0.206) (0.069 to 0.083) (0.043 to 0.057) (1.558 to 1.839) (1.195 to 1.295) (1.623 to 1.841) (31.3 to 39.2) (76.7 to 89.9) (39.9 to 73.7) (19.0 to 47.3) (10.8 to 25.2) (7.8 to 17.6) (2.64 to 7.52) (0.00 to 0.22) (11.2 to 25.9) (6.5 to 16.9) (0.16 to 1.78) (8.3 to 85.0) (0.0 to 57.2)
(0.155 to 0.192) (0.047 to 0.054) (0.023 to 0.029) (1.669 to 1.801) (1.317 to 1.399) (1.791 to 1.935) (31.9 to 36.1) (74.1 to 81.6) (38.35 to 60.0) (22.4 to 39.9) (11.5 to 22.9) (7.0 to 14.7) (2.81 to 7.76) (0.00 to 0.00) (12.5 to 28.4) (8.6 to 20.0) (0.01 to 0.68) (10.8 to 67.8) (0.0 to 52.3)
(0.154 to 0.188) (0.047 to 0.063) (0.023 to 0.040) (1.741 to 1.891) (1.330 to 1.583) (1.803 to 1.963) (29.2 to 36.1) (59.8 to 81.6) (31.2 to 60.0) (19.4 to 39.0) (11.4 to 22.6) (6.8 to 14.25) (2.73 to 7.08) (0.00 to 0.03) (13.7 to 28.7) (9.9 to 21.1) (0.07 to 1.33) (10.8 to 70.0) (0.00 to 56.0)
Both p values are comparisons between MCO and standard IMRT. Shown are the median, first, and third quartiles. Wilcoxon signed rank tests performed for each with significance level at p ¼ 0.05. Significant p-values are indicated in bold. CI ¼ conformity index; R ¼ rectum; B ¼ bladder; P_B ¼ penile bulb. Subscripts show percentage dose point that volume on DVH is analyzed.
Oncentra PB algorithm and RayStation CC algorithm, the average gammas were 0.24 ⫾ 0.15 and 0.22 ⫾ 0.12, respectively. Plots of average rectal dose against HI for both the calculation algorithms for all the patients showed minimal differences across the entire Pareto front (data not shown). Table 2 summarizes dosimetric data attained from the DVH of the conventional IMRT plan and the MCO plans that satisfied CHHiP requirements and had minimum rectum dose for each patient. The tabulated data show that the target volumes of the MCO plans were more homogeneous than those of the conventionally optimized plans for all 3 PTVs. The average rectal dose and rectal dose points were also lower for the MCO plans. The pattern was reversed if comparing the planned dose to the bladder and dose conformity to the target volumes, where better results were observed for the conventionally optimized plans. Conformity was not statistically different for dose surrounding all the PTVs (as depicted by PTV 1), although differences were observed for the boost regions. The clinically acceptable MCO plans had lower rectal doses for almost all tolerance levels and for the dose to 100% of the bladder volume. The dose received by 81% and 68% of the bladder volume was significantly less for the conventionally optimized plans. The best possible clinically acceptable MCO plans, irrespective of the navigated plan, followed a similar trend, although the average dose to the rectum reduced further. Pareto front analysis Figure 2 shows Pareto fronts of target homogeneity vs average rectal dose for all navigated plans (fluence space) of each patient plotted along with the same plans, after they had been converted to deliverable MLC apertures and accurate dose calculated with CC. In the region of the Pareto front where the average rectal dose was largest, the difference between the navigated and deliverable plan was minimal although the difference progressively increased as the average rectal dose reduced. The navigated Pareto front is closest to the axis, which would suggest that this plan set is of superior quality compared with the deliverable plan. Figure 3 shows a representative example of a DVH distribution for a clinically acceptable MCO plan (patient 10) both in the navigated state (solid line) and after a conversion into deliverable MLC segments with the final dose calculated with CC (dashed line).
Differences between the DVH plot of PTV 3 are minor overall, but nevertheless relevant because they occur close to the dose received by 99% of the target volume, which can have a significant effect on target coverage when viewing dose distributions. The differences between the navigated and deliverable DVH plots of the rectum are of larger magnitude. RayStation permits the user to select a scalarvalued target priority weight that determines how replication of the navigated DVH in target structures is traded against replicating DVH curves corresponding to healthy tissue. The dashed lines indicate deliverable DVHs obtained with a target priority of 30, which is the system default and the setting used for all numerical results presented in this study. The sensitivity with respect to this parameter is illustrated by the dotted DVHs that correspond to a deliverable plan generated with a target priority of 50. As expected, a higher priority to targets results in improved target coverage but at the expense of large deviations for the critical structures. A final assessment was performed regarding to what extent clinical decisions can be made on the basis of navigated dose distributions alone. This analysis showed that the navigated plans satisfying CHHiP requirements and having minimum rectum dose rarely translate to the acceptable deliverable plans following a final dose calculation using RayStation or Oncentra. The acceptable navigated plan with minimum rectum dose could be converted to an acceptable deliverable plan using RayStation CC algorithm for 7 of 10 patients. Also, the acceptable navigated plan with minimal rectum dose could be converted to an acceptable deliverable plan on Oncentra PB algorithm for 5 of 10 patients. On a number of occasions navigated plans were not clinically acceptable owing to either lack of target coverage or OAR tolerances not being met but were, in fact, acceptable following conversion to a deliverable plan. The best deliverable column in Table 2 shows plans satisfying CHHiP requirements and having minimum rectum dose irrespective of the acceptability of the initial plan navigation. There was a clinically acceptable plan with a lower rectal dose in 6 of 10 patients if the plan exactly matched the navigated plans compared with plans that required a clinically acceptable navigated plan. Discussion In this study, we have validated RayStation MCO against Oncentra standard IMRT optimization. MCO and standard IMRT
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]]
Fig. 2. Pareto fronts of 10 patients comparing multicriteria optimization (MCO) navigation (MCO Navigated, plan following segmentation and collapsed-cone final dose calculation (MCO deliverable, , solid line).
▲
5
△, long dashed line) with the corresponding deliverable MCO
6
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]]
Fig. 3. DVH of patient 10 with MCO navigated plan (solid) and the final deliverable plan after collapsed-cone final dose calculation optimized with target priority 30 (dashed) and 50 (dotted). (Color version of figure is available online.)
plans were observed to be of similar quality. The MCO plans also showed similar or superior target homogeneity compared with the conventionally optimized IMRT plans. Rectal sparing of the MCO plans were to some extent at a cost of less conformal dose distributions and higher bladder doses. The analysis of the dose deviations between navigated and final deliverable MCO plans revealed that sacrifices in dose quality owing to the conversion was primarily made in terms of sparing of critical structures. In addition, these dose differences were found to increase as a higher priority was placed on avoidance of the rectum. Differences between the final dose calculation methods of the 2 independent planning systems used in this study (i.e., PB in Oncentra and CC in RayStation) only resulted in very minor perturbations of the Pareto fronts, and these differences were also confirmed to be minor in experimental measurements of the accuracy of delivery. Slight improvements in plan quality would be demonstrated if the navigation to delivery step was eliminated. Inverse planning for IMRT delivery is well known to often require a significant amount of manual parameter tuning before a final clinically acceptable plan is settled upon by the planner. This plan may then be checked by the clinician who may then ask for further refinement depending on the quality of the plan presented. MCO can significantly improve the efficiency of this process as the database of possible plans can be presented to and navigated by the clinician before settling on a specific plan.11,26 It is clear from this study that MCO plans of similar quality to standard IMRT plans can be created for patients with prostate cancer where stringent tolerances are often used. The flexibility achieved by having the option of navigating to the required plan by a clinician can also undoubtedly improve planning efficiency. On the negative side, our results show that the dose deviations that occur because of the conversion of a navigated treatment plan into deliverable machine setting in some cases can render an initially acceptable treatment plan unacceptable for clinical delivery. Such dose deviations would in a clinical setting need to be counteracted by parameter adjustments followed by generation of another deliverable treatment plan; a process that defeats the initial purpose of MCO. In this study, the main dose deviation because of the conversion process resulted in either lack of target coverage or hotspots in the targets. Significant increases were also observed in the dose to critical structures. The Pareto fronts analysis in Fig. 2 clearly indicates that there is potential for the navigated plans to better approximate the
deliverable frontier as there appears to be a systematic shift of the deliverable curve from the navigated curve, whereby the differences get larger at lower average rectal doses. This result was consistent in all the 10 patients. Further analysis also showed that the navigated plan satisfying CHHiP requirements and having minimum rectum dose may not result in an acceptable deliverable plan. For a number of patients more acceptable MCO plans could have been utilized but because the navigated plans were unacceptable they would have been discounted by the clinician. Better characterization of the navigated plans would constitute a solution to this problem. Prostate plans are not the most complex IMRT plans to create,19 although placing a high priority on avoidance of the rectum does increase the complexity of the plan. It appears from the data presented in this study that the more complex initial navigated plans (with lower average rectal dose) deviate most from the final deliverable plans. Hence, it may be important to validate any new approximations, developed for use with MCO planning, using both the multidose level IMRT treatment on a cohort of patients with prostate cancer presented in this article, as well as more complex cases, such as patients with head and neck cancer. Further work is underway through improved algorithms to generate base or navigated plans that correlate more closely with deliverable plans. Examples of such efforts include navigation among segmented plans where each plan has unique MLC shapes and restrictions are enforced regarding how plans can be combined,13,14,27 and unrestricted navigation among segmented base plans that have shared MLC shapes.22 It is important that these are clinically validated, and these should then enhance the current power of MCO planning.
Conclusions MCO appears to have the potential to create equivalent or better plans for patients with prostate cancer, than those that are created using standard inverse planning methods. Further improvements in MCO planning may be possible through better approximation of deliverable plans during the navigation phase of MCO planning.
Acknowledgments The authors thank Björn Hårdemark, RaySearch Laboratories, for his helpful comments on the article. References 1. Sheets, N.C.; Goldin, G.H.; Meyer, A.M.; et al. Intensity-modulated radiation therapy, proton therapy, or conformal radiation therapy and morbidity and disease control in localized prostate cancer. J. Am. Med. Assoc. 307(15):1611–20; 2012. 2. Bohsung, J.; Gillis, S.; Arrans, R.; et al. IMRT treatment planning:- a comparative inter-system and inter-centre planning exercise of the ESTRO QUASIMODO group. Radiother. Oncol. 76(3):354–61; 2005. 3. Chung, H.T.; Lee, B.; Park, E.; et al. Can all centers plan intensity-modulated radiotherapy (IMRT) effectively? An external audit of dosimetric comparisons between three-dimensional conformal radiotherapy and IMRT for adjuvant chemoradiation for gastric cancer Int. J. Radiat. Oncol. Biol. Phys. 71(4):1167–74; 2008. 4. Seco, J.; Clark, C.H.; Evans, P.M.; et al. A quantitative study of IMRT delivery effects in commercial planning systems for the case of oesophagus and prostate tumours. Br. J. Radiol. 79(941):401–8; 2006. 5. Shepard, D.M.; Earl, M.A.; Li, X.A.; et al. Direct aperture optimization: a turnkey solution for step-and-shoot IMRT. Med. Phys. 29(6):1007–18; 2002. 6. Craft, D.; Halabi, T.; Bortfeld, T. Exploration of tradeoffs in intensity-modulated radiotherapy. Phys. Med. Biol. 50(24):5857–68; 2005. 7. Hoffman, A.; Siem, A.; den Hertog, D.; et al. Derivative-free generation and interpolation of convex Pareto optimal IMRT plans. Phys. Med. Biol. 51 (24):6349–69; 2006. 8. Craft, D.; Halabi, T.; Shih, H.A.; et al. An approach for practical multiobjective IMRT treatment planning. Int. J. Radiat. Oncol. Biol. Phys. 69(5):1600–7; 2007.
C.K. McGarry et al. / Medical Dosimetry ] (2014) ]]]–]]] 9. Serna, J.; Monz, M.; Küfer, K.-H.; et al. Trade-off bounds for the Pareto surface approximation in multi-criteria IMRT planning. Phys. Med. Biol. 54(20): 6299–311; 2009. 10. Hong, T.S.; Craft, D.L.; Carlsson, F.; et al. Multicriteria optimization in intensitymodulated radiation therapy treatment planning for locally advanced cancer of the pancreatic head. Int. J. Radiat. Oncol. Biol. Phys. 72(4):1208–14; 2008. 11. Craft, D.L.; Hong, T.S.; Shih, H.A.; et al. Improved planning time and plan quality through multicriteria optimization for intensity-modulated radiotherapy. Int. J. Radiat. Oncol. Biol. Phys. 82(1):e83–90; 2012. 12. Dearnaley, D.; Syndikus, I.; Sumo, G.; et al. Conventional versus hypofractionated high-dose intensity-modulated radiotherapy for prostate cancer: preliminary safety results from the CHHiP randomised controlled trial. Lancet Oncol. 13:43–54; 2012. 13. Salari, E.; Unkelbach, J. A column-generation-based method for multi-criteria direct aperture optimization. Phys. Med. Biol. 58(3):621–39; 2013. 14. Craft, D.; Richter, C. Deliverable navigation for multicriteria step and shoot IMRT treatment planning. Phys. Med. Biol. 58(1):87–103; 2013. 15. Philips, M.; Holdsworth, C. When is better best? A multiobjective perspective Med. Phys. 38(3):1635–40; 2011. 16. Ottosson, R.O.; Engstrom, P.E.; Sjöström, D.; et al. The feasibility of using Pareto fronts for comparison of treatment planning systems and delivery techniques. Acta Oncol. 48(2):233–7; 2009. 17. Petersson, K.; Ceberg, C.; Engström, P.; et al. Beam commissioning and measurements validating the beam model in a new TPS that converts helical tomotherapy plans to step-and-shoot IMRT plans. Med. Phys. 38(1):40–6; 2011. 18. Pardo-Montero, J.; Fenwick, J. Tomotherapy-like versus VMAT-like treatments: a multicriteria comparison for a prostate geometry. Med. Phys. 39(12):7418–29; 2012.
7
19. McGarry, C.K.; Chinneck, C.D.; O’Toole, M.M.; et al. Assessing software upgrades, plan properties and patient geometry using intensity modulated radiation therapy (IMRT) complexity metrics. Med. Phys. 38(4):2027–34; 2011. 20. Bokrantz, R.; Forsgren, A. An algorithm for approximating convex Pareto surfaces based on dual techniques. INFORMS J. Comput. 25(2):377–93; 2013. 21. Hårdemark, B.; Liander, A.; Rehbinder, H.; et al. Direct machine parameter optimization with RayMachines in Pinnacle3s. White paper, RaySearch Laboratories, Stockholm, Sweden; 2003. 22. Bokrantz, R. Multicriteria optimization for volumetric-modulated arc therapy by decomposition into a fluence-based relaxation and a segment weight-based restriction. Med. Phys. 39(11):6712–25; 2012. 23. Bortfeld, T.; Schlegel, W.; Rhein, B. Decomposition of pencil beam kernels for fast dose calculations in three‐dimensional treatment planning. Med. Phys. 20 (2):311–8; 1993. 24. Mott, J.H.; Livsey, J.E.; Logue, J.P. Development of a simultaneous boost IMRT class solution for a hypofractionated prostate cancer protocol. Br. J. Radiol. 77 (917):377–86; 2004. 25. McGarry, C.K.; McMahon, S.J.; Craft, D.; et al. Inverse planned constant dose rate volumetric modulated arc therapy (VMAT) as an efficient alternative to fivefield intensity modulated radiation therapy (IMRT) for prostate. J. Radiother. Pract. 13(1):68–78; 2014. 26. Wala, J.; Craft, D.; Paly, J.; et al. Maximizing dosimetric benefits of IMRT in the treatment of localized prostate cancer through multicriteria optimization planning. Med. Dosim. 38(3):298–303; 2013. 27. Fredriksson, A.; Bokrantz, R. Deliverable navigation for multicriteria IMRT treatment planning by combining shared and individual apertures. Phys. Med. Biol. 58(21):7683–97; 2013.