The criticism on the PTV concept in lung SBRT arises from the notion that its combination with type–B and Monte Carlo (MC) dose optimization would result in excessive and inconsistent GTV dose owing to an artificial increase of photon fluence in the low density lung tissue. Such limitation of current SBRT practice is also recognized in the recent ICRU report 91on prescribing, recording, and reporting of stereotactic treatments with small photon beams. This report further suggested that robust optimization and GTV–based prescriptions are viable options to possibly improve the overall dosimetric accuracy and quality while reducing the variability in dose reporting and potentially even in dose prescription.
By analyzing the dosimetric variability and robustness resulting from two common PTV–based and two other worse case–based robust optimization methods, this study is now able to provide more clarifications to the pitfalls of PTV concept in lung SBRT. Additionally, by analyzing further the dosimetric results by different dose prescription methods according to the ICRU recommended coverage prescription and GTV median dose prescription, we identified the dominant factor that contributes to the variability of GTV dose.
PTV– and WCS–based SBRT using ICRU 91 recommended coverage prescription
SBRT plans optimized and prescribed to the PTV resulted in significant overexposure to the GTV compared to those plans optimized for WCS as expected. The GTV receives much higher dose, with the GTV median dose
D50 about 17 and 22% over the prescription dose for ITV–based and mid–ventilation based PTV optimizations, respectively. Although higher dose to the GTV is generally not a concern and even desired for SBRT, part of the excessive dose is in effect burdened by the surrounding normal organs including the normal lung and chest wall that are encompassed in the PTV. For lesions that are close to the chest wall, the volume receiving ≥30 Gy (V
30) was reduced significantly by up to 29.5 cm
3 (74%) and 31.4 (73%) by using WCS optimization on the averaged intensity projection (AIP) image to the ITV and on all 4DCT images directly to the GTV in comparisons to the conventional PTV approach based on the ITV. The dosimetric benefit of WCS optimization to limit the chest wall dose was also reported by Zhang et al. [
21]. In their study, 8 of 20 patient plans optimized and prescribed to PTV showed chest wall dose above the limit whereas all WCS plans optimized to the ITV fulfilled the dose constraint. In this study, we showed that WCS optimized to the GTV can further improve the chest wall dose.
Besides the dosimetric inferiority to WCS optimization, the other major pitfall of PTV concept for plan optimization is that inconsistent GTV doses between individual patients (i.e., inter–patient variability) occur even with the same PTV prescription. However, our results clearly demonstrated that inconsistent GTV dose is not unique to the PTV concept. Other methods that avoid the PTV concept in SBRT planning equally suffer from inconsistent GTV doses. Specifically, robust optimization that replaces the PTV concept by the worst case method also shows inconsistent GTV dose. This was evidenced by the equivalent variances of GTV
D98, D50 and
D2 among all PTV and WCS–optimized plans (Table
1). In principle, one would expect zero or minimal variability of GTV
D98 at and close to the prescription point of GTV
D98 or ITV
D98 in the WCS–optimized plans. Recall that robust optimization in this study was implemented to ensure the prescription dose in the worst case scenario, that is, the GTV
D98 was optimized to equal to or at least 54 Gy in the worst case scenario but it could be any values > 54 Gy in other scenarios. Since the nominal scenario does not necessarily coincide with the worst case scenario, and in fact hardly does, GTV
D98 does not necessarily arrive exactly at 54 Gy in the nominal scenario and hence variability. On the other hand, any renormalization made to equalize GTV
D98 to 54Gy in the nominal scenario would invalidate the plan robustness that was achieved to ensure the prescription dose for the worst case scenario.
When Lacornerie et al. [
11] initially argued against type–B dose engine for dose optimization using the PTV concept, they claimed “the GTV will be overexposed when it moves into the regions with increased photon fluence” but without providing results to assess the magnitude of the matter. Following this line of argument, if type–B dose engine did induce excessive photon fluence in the low density PTV border one would expect the dose received by the GTV to be higher in other respiratory phases than in the planning phase. We therefore followed the phase–to–phase changes in the GTV doses. Our results show that all GTV dose parameters, except for
D2 using the mid–ventilation concept, were statistically equal among the ten 4DCT images for the PTV–optimized plans. Guckenberger et al. [
24] previously optimized for the PTV coverage
D95 on the end–exhale CT, in which case the type–B dose engine would in principle drive the optimizer to deposit the maximal fluence at the opposite end–inhale position. Interestingly, the authors found no significant GTV dose differences when these plans were recalculated on the end–inhale CT. Maximum differences of 6.9 ± 3.1% and 2.4 ± 1.8% for GTV
D99 and
D50 were reported, respectively. This study observed smaller maximum differences of 2.7 ± 1.4% and 0.9 ± 0.5% for GTV
D98 and
D50, respectively. The discrepancy is presumably attributed to the different planning CT datasets (end–exhale vs. AIP images) for which the fluence optimization were carried out.
Here, we attempt to offer an explanation to the negligible GTV dose difference among breathing phases from the principles of conventional radiotherapy and SBRT. In conventional VMAT–based radiotherapy, a uniform dose profile (e.g., +/− 5%) across the PTV is often demanded and achieved by a fluence profile that is typically characterized with horns at the PTV edge to compensate for the beam penumbra. Thus, the GTV may experience an increase of fluence when it moves towards the PTV border. The magnitude of this fluence horn increases from water density to lung density to counterbalance the deteriorating condition of charged particle equilibrium. By contrast, SBRT allows higher dose in the tumor center (as much as 167% when normalized to the maximum dose at 60% on the PTV surface). In this case, the “horn” effect diminishes as the demand of photon fluence is counterbalanced by the allowed lower dose to the region around the PTV edge. The other possible reason could be that commercial planning system generally switches the type–A dose engine to type–B dose engines only at certain steps for fluence correction during the dose optimization and in final dose calculation.
Additionally, we examined the variances of different GTV dose parameters among the ten respiratory phases. Our hypothesis is that if type–B dose engine did drive up the photon fluence in the PTV–optimized plans the inter–phase variability of these GTV dose parameters would become significantly different. This hypothesis is based on the fact that individual patients have different characteristics (e.g., tumor size, motion amplitude, lung density, etc) and hence the extent to which the photon fluence were to be driven up would vary substantially. When the GTV moves in different spatio–temporal positions of the respiratory cycle it would receive photon fluence of varying degree from phase to phase that is patient dependent. Nonetheless, we found that both PTV and WCS optimizations resulted in equal variances of all GTV dose parameters among the ten respiratory phases. Interestingly, the inter–quartile ranges (IQR) of GTV
D98 resulting from WCS optimized plans using all 4DCT images were found to be more variable than from other PTV–optimized plans. This large but insignificant variability of GTV
D98 is hypothesized to have originated from the specific worst case optimization method. Compared to the voxel–wise and objective–wise robust methods, the composite worst case method implemented by the RayStation planning systems behaves to maximally minimize the objective value on the worst case scenario at the cost of higher objective values and thus larger dosimetric fluctuation in many other possible scenarios [
25]. Since the worst case scenario may correspond to different breathing phases with different patient characteristics, relatively large variability of
D98 among breathing phases was observed. Nonetheless, by WCS optimization, particularly using all 4DCT images, the highest robustness was achieved to prevent the dose limits in the normal tissues from being exceeded when the target is displaced into different respiratory positions.
As the final validation, we compared the optimized dose on a single CT and the recalculated doses summed over all 4DCT images. Such comparisons offer clarifications to two important issues concerning the non–consistency of PTV concept in lung SBRT. Firstly, if type–B dose engine induced excessive fluence in PTV–based optimization, the GTV would eventually accumulate significant higher dose when it moved into different breathing phases. However, no clear indication of overexposure to the GTV can be associated with PTV–based optimization (Table
3). The GTV
D50 and
D2 obtained from PTV–optimized plans for the ITV and mid–ventilation concepts changed by 0.3 Gy only after dose summation and on the contrary decreased rather than increased. The significant increase of GTV
D98 in the PTV–optimized plans based on the ITV concept does not appear to be related to the type–B dose engine because it did not occur to the other PTV–optimized plans that adopted the mid–ventilation concept. Instead, it was presumably caused the systematic change in using the AIP images for dose optimization to the mid–ventilation images for dose accumulation. For the rather extreme situation using the end–exhale CT for fluence optimization, neither did Guckenberger et al. [
24] observe serious problem of excessive build up of photon fluence at the opposite end–inhalation that caused a significant change in the overall GTV dose either. More interestingly, the authors too found an increase rather than a decrease in the summed GTV dose (presumably
D95) by less than 1% or 0.7 Gy only. Among all GTV dose parameters,
D50 appears to be the most robust against changes showing no statistical significance except for the ITV–based robust optimization. Based on these results, we conclude that type–B dose engine, per se, does not significantly increase the GTV dose. The significantly higher GTV dose in the PTV–optimized plans than WCS optimized plans is rather a direct consequence of the prescription method.
Secondly, equal variances of the GTV dose parameters among the PTV and WCS–optimized plans are still observed after dose summation over the ten 4DCT images. The inter–patient variability (one standard deviation) changes only by 0.1 Gy after dose summation in all but the GTV
D98 of the WCS–optimized plans (0.9 Gy). This simply means that the inconsistency of GTV dose cannot be easily resolved by migrating from the PTV concept to robust optimization irrespective of the type–B dose engine [
1,
14]. For the same reasoning, we would argue that using two classes of dose engines, a type–A for fluence optimization followed by a type–B for subsequent dose calculation and renormalization will not resolve the inconsistent GTV dose either. We would further argue that PTV concept, in its very design to account for geometric uncertainty, shall not be considered as a pitfall. Consistency of clinical outcome report shall not be compromised provided that the advanced dose engines are used to estimate and report the GTV dose parameters following the ESTRO ACROP recommended guidelines [
11].
PTV– and WCS–based SBRT by GTV median dose renormalization
Lebredonchel et al [
14] suggested that prescribing based on 50% mass of the PTV can somewhat stabilize variability of the target dose but they concluded further that moving away from the PTV concept for prescription remains the only solution if using type–B dose engine. They came to this conclusion because the GTV median dose
D50 differs substantially with variable lung density and tumor size when prescription is done to the PTV. However, this conclusion is considered as partly true only because our results already showed that other PTV–free concept by the worst case method does not stabilize the target doses either when the ICRU recommended prescription by coverage (i.e., GTV
D98 or ITV
D98) was followed. Instead, the prescription method has the major impact on the variability of GTV dose. After renormalization based on GTV
D50, the separations of the DVH families became much packed together for all plans optimized using different concepts (Fig.
4), as compared to those obtained from prescription by coverage (Fig.
1). The resulting SDs of
D98 and
D50 and
D2 are limited to 1 Gy for PTV– and 1.4 Gy for WCS–optimized plans, respectively. Focusing on the concept of ITV as motion encompassing, Lang et al. similarly showed that the SDs of PTV
D98 and
D50 and ITV
D98 of 38 patients are limited to 1.5 Gy after ITV
D50 renormalization to 57 Gy [
18]. They also showed that the ITV
D50 renormalization is superior to renormalization by ITV/PTV coverage
D98 as it can reduce the variability of PTV and ITV dose parameters among delivery techniques (dynamic conformal arc vs. VMAT). More importantly, the differences of GTV
D98 and
D50 and
D2 among optimized plans based on the PTV concept and the WCS method (Table
4) were found to reduce markedly. These results are still valid despite the variation of tumor position in the respiration cycle, with GTV
D50 being the only dose parameter that showed statistically significant difference. However, the absolute difference of 0.2 Gy is deemed clinically unimportant. Same as the results of coverage prescription, the median dose turned out to be the most robust against uncertainty of tumor position among other GTV dose parameters.
The effect of GTV
D50 renormalization is also marked at the phase to phase level (Fig.
5). The median of all GTV dose parameters became much closer among the plans that adopted different concepts for setup and motion compensation. Compared to the prescription by coverage method recommended by ICRU 91 report, the maximum inter–phase difference of GTV
D98 was reduced by 2.4, 4.8 and 2.4% and 1.0% for PTV optimization by the ITV and mid–ventilation concepts, and WCS optimization to the ITV and GTV, respectively.
In summary, when SBRT plans are directly prescribed or renormalized to the GTV median dose
D501.the consistency of GTV dose across the near–minimum, median, and near–maximum points is significantly improved, i.e. reduced inter–patient variability
2.harmonization of GTV dose is made possible for lung SBRT plans that adopt different concepts to handle geometric uncertainty caused by respiratory motion.
The first point simply implies that one can continue with the PTV concept for dose planning. The second point implies that consistent GTV dose shall be ensured between SBRT centers employing either the PTV concept or the worst case scenario concept in dose planning, and different delivery techniques as indicated by Lang et al. [
18].
On the other hand, one may question the value of robust optimization concerning its computational overheads, if by
D50 GTV prescription alone can simply harmonize the GTV dose among optimization solutions. From the normal tissue dose perspective, our phase–by–phase analysis indicates that WCS optimization in general improved the dosimetric robustness, resulting in the fewest number of dose deviations from the OAR limits. Furthermore, lower NL
V5 and MLD (Table
3) during respiration were constantly observed in the WCS optimization group regardless of the prescription method. In particular, WCS optimization to the GTV using all 4DCT images resulted in the lowest normal tissue dose and highest robustness against deviation of normal tissue dose limit among all optimization methods.
Limitation of the study
This study was designed by assuming the same amount of geometric uncertainties from tumor motion and patient setup in the calculation of the PTV and in the definition of the WCS parameters. Nonetheless, our results considered exclusively the uncertainty of tumor position due to breathing motion. The validity of our results shall hold because uncertainty of respiratory motion, which is considered as systematic in our phase–to–phase analysis for the GTV dose changes, is much greater than that of setup limited to millimeter accuracy with stereotactic image guidance.
The other limitation is the small number of patients which may subject our results to bias. Only 2 out of 13 patients showed tumor motion more than 1 cm. It is unclear whether our dosimetric results will remain unchanged if more patients with larger amplitude of tumor motion are included.
We also acknowledge that the exact formulation of the robustness optimization may have an influence on the dosimetric results [
22]. Despite the numerous robustness optimization algorithms, there is only one commercial planning system that makes robust optimization available for clinical use. This study, like many other previous ones, was based on the worst case scenario optimization from the same planning system. Lastly, this study focused on a certain type (convolution–superposition) and class (type–B) of dose engine. Systematic difference between Monte Carlo and type B dose engines is well known especially in cases where extreme electron charged disequilibrium exists [
26]. Further investigation with Monte Carlo dose engine is warranted to generalize the present findings.