Background
Dose prescription in radiation oncology is one of the most fundamental considerations during the treatment planning process. Prescribed doses for a given clinical scenario are based on experience, historical precedent, and the results of clinical studies, including phase I clinical trials [
1]. For the most part, especially for common tumors treated in various contexts (adjuvant or definitive), dose-fractionation schemes are reasonably well established and commonly employed. A stated dose-fractionation scheme, however, tells an incomplete story. Meeting organ at risk (OAR) dose constraints may lead to various compromises in tumor coverage [
2]. In addition, doses well above the prescription dose (hot spots) may be delivered to sub-volumes within the tumor [
3].
When there are no critical normal tissues ensconced within a tumor mass, hot spots most likely will only increase tumor control probability (TCP), without anticipated worsening of toxicities, and can additionally aid in achieving steep dose gradients outside of the target [
4‐
7]. Radiosurgery treatment plans are a prime example of this approach to treating cancer with radiation. Based in considerations of beam penumbra, radiosurgery plans often prescribe dose to very low isodose clouds (e.g., 50–80%) at the margins of tumors, leading to optimal gradient of dose outside of the target, and, simultaneously, extreme hot spots within the tumor [
8,
9]. With intensity modulated radiotherapy (IMRT) planning, relaxing homogeneity constraints in the target (allowing hot spots, although not purposely planning for them) can also improve OAR sparing [
7,
10]. Increasing the magnitude and spread of hot spots within tumors should improve the TCP from fundamental tenets of radiation biology. This may be especially true if there is heterogeneity of tumor cell radiosensitivity (relating, for example, to repair capabilities or hypoxia) throughout the tumor mass [
5].
Thus, two radiation plans that are predicted to be isotoxic because they both meet normal tissue constraints may 1) not be equivalent with respect to the TCP and 2) in fact not be isotoxic, as our knowledge of normal tissue dose-fractionation effects is incomplete, and the plan that has the sharper gradient of dose may ultimately be associated with a lower risk of normal tissue complication. To improve the TCP, the planner can purposely “pack” or embed high dose within a tumor target as a planning objective during the IMRT optimization process. This approach can be considered a special/non-traditional simultaneous integrated boost (SIB) planning approach with focus on internal boosting
within the gross tumor. Such a concept has been used previously in Gamma Knife treatment planning [
11]. While intentionally embedding high doses in tumors using IMRT is possible, it may be achieved using multiple optimization approaches which may not be equally effective with respect to dose escalation internal to the tumor and dose gradient external to the tumor [
12,
13]. To the authors’ knowledge there has not been a comparison of optimization approaches in terms of maximum embedded doses achievable while maintaining a given dose to the tumor margin and maintaining a high dose constraint outside of the target.
In this report, we consider five approaches to this problem- Approach 1 involves the definition of an embedded dose objective within the target, without specifying its physical location, using a dose-volume histogram (DVH) objective. Approach 2 involves the creation of a defined sub-volume within the target to be treated with an integrated/embedded boost. This approach is already in use in the clinic in various forms [
13‐
15]. Approaches 3, 4 and 5 involve maximizing the generalized equivalent uniform dose (gEUD) in the target using a biological modelling-based objective function [
16]. These gEUD approaches have been proposed in previous studies [
12]. We compared these approaches by optimizing co-planar VMAT plans for a series of spherical targets with varying diameters and analyzing the resultant dose distributions in terms of embedded hot spots and gradient outside the target. We also analyze the machine delivery parameters of each plan to characterize the mechanism leading to differences in dose between approaches, and implemented all approaches in a single patient example to investigate clinical feasibility with true patient anatomy and dose constraints.
Discussion
This planning study compared five optimization approaches intended to embed high doses of radiation in tumors, thereby creating plans with extreme dose heterogeneity. While heterogeneous dose distributions are common in radiosurgery and stereotactic body radiotherapy (SBRT), this has typically been a byproduct of increasing the dose gradient at the tumor edge to improve normal tissue sparing, rather than purposeful embedding of hot spots within the tumor to improve TCP. This study demonstrates various optimization approaches to embed high doses within tumors, and demonstrates that there is minimal trade-off between increased tumor dose and decreased normal tissue dose with average increases in R50% ≤11%. Other approaches to embedded dose escalation beyond those studied here have also been reported, including the use of high-dose-rate (HDR) brachytherapy dose distributions as a guide for planning [
22,
23]. It should be noted that embedded boosting differs from traditional SIB planning, which typically involves treating areas at risk for microscopic disease to a certain intermediate dose and areas of gross disease to a higher dose. Embedded boosting focuses on internal dose escalation
within gross disease. The approaches investigated in this study maintained a minimum dose specification for the target indicated as the prescription dose, consistent with ICRU 83 level 2 dose reporting; however, central target doses significantly exceeded this dose level and could also be reported using maximum PTV dose or D2% to maintain an accurate record of the dose distribution [
3].
Previous work has demonstrated that decreasing aperture sizes and increasing MUs provide increased target dose while maintaining dose outside of the target for radiosurgery [
24,
25]. Tanyi et al. demonstrated that using a negative MLC margin of 1 mm resulted in superior TCP compared to zero or positive MLC margins for intracranial lesions treated with conformal arcs [
24]. While increasing MUs is associated with increased OAR and integral dose, this effect is counter-acted by decreasing MLC separation which decreases beam overlap outside of the target. Similarly, our results show that the plan with the highest MUs and smallest mean MLC separation provided the highest embedded target dose. This plan was produced using the DVH-based optimization approach in both the phantoms and patient example. Alternatively, the gEUD-based optimization approach assuming a = − 5 provided the highest gEUD values of all optimization approaches. The relative impact of the high-dose regions provided by the DVH-based approach and the higher gEUD provided by the gEUD-based approach on TCP is unclear. The relative impact on TCP may depend on the distribution of tumor cell radio-sensitivity within the target [
4,
5].
The DVH approach provided the highest embedded target doses compared to the other approaches for two potential reasons. 1) The DVH approach provides the optimizer with flexibility in terms of the size and shape of the embedded region of escalated dose, creating a larger viable solution space than the sub-volume approach leading to an improved solution. 2) The DVH approach directly emphasizes increased dose to a small sub-volume of the target, whereas the gEUD approaches emphasize increased dose to the entire target. In fact, the negative values of the gEUD volume-effect parameter
a specifically emphasize the reduction of low-dose regions, rather than the creation of high-dose regions [
16]. For these reasons, the DVH approach provided the most flexible and direct way to embed regions of high dose in targets. In this study, the DVH and sub-structure approaches were implemented using a 10% sub-volume of the PTV to enable initial comparison between approaches; however, this specific sub-volume value could be further optimized to improve results beyond those achieved in this study.
The minimum target dose objectives used for each of the five experimental planning approaches were selected to be as high as possible while still allowing the optimizer to either converge or complete 60 iterations without error. When objectives were selected that were higher than permissible, the treatment planning system would indicate that no further optimization could be performed after 10 iterations, returning a sub-optimal plan. This behavior was attributed to the instance of the SmartArc optimization algorithm employed in this study, which makes use of iterative gradient descent, and is therefore sensitive to initial conditions and is susceptible to local minima potentially leading to sub-optimal plans [
18]. Optimizers available in other treatment planning systems may enable an increased range of dose objectives; however, we expect that the relative performance of the objective functions compared in this study to be applicable to other treatment planning systems making use of similar gradient descent-based optimization.
We also performed a limited investigation of the impact of the minimum D10% objective on the maximum achievable target dose for the DVH and sub-structure optimization approaches for the 5 cm target diameter, and found that increasing objective values > 220% of the desired tumor margin dose had little impact on the resultant region of escalated dose. We were able to use minimum D10% objectives > 220% of the tumor margin dose for all target diameters when using the DVH approach, and for target diameters ≤5 cm when using the sub-structure approach. We were able to use minimum D10% objectives of 200% of the desired tumor margin dose for the 6, 7, and 8 cm target diameters using the sub-structure approach. A minimum D10% objective of 200% of the desired tumor margin dose may be a pragmatic starting point when using the DVH or sub-structure approach. Although we have not yet performed a similar analysis for the gEUD optimization approaches, we were able to use minimum gEUD objectives > 220% of the tumor margin dose for all target diameters, which we expect to be within the plateau region of resultant target D0.1 cc.
The oncologic advantages of embedded hot spots within a tumor, and the relationship to the magnitude of the hot spots, are uncertain. Hot spots emerge naturally from treatments such as Gamma Knife radiosurgery and interstitial brachytherapy. Modeling studies have shown the ability of intra-tumoral boosts to increase tumor control probability [
4‐
6]. Embedded hot spots would likely be especially beneficial if tumor cell radioresistance is heterogeneously distributed throughout the tumor [
5]. A variety of clinical studies have investigated the relationship of peripheral tumor dose as well as internal hot spots to tumor control, with some reports showing an association between internal dose escalation and better tumor control outcomes [
26,
27]. Ideally the areas of a tumor which contain the most resistant clonogens could be identified and selected for specified internal boost, and there is much interest in the use of imaging studies to identify tumor sub-volumes for boosting, but such information is often lacking from available imaging studies, and may in fact be fluid depending on, for example, oxygenation patterns during a radiation treatment course [
28,
29]. As our knowledge of the impact of functional imaging studies on guiding embedded dose escalation evolves, our perspective is the following: for selected tumors which lack critical normal structures interspersed within the tumor, and for a given dose prescribed to the margin of the tumor, embedding high doses (beyond the tumor margin dose) within the tumor is likely to be beneficial, relative to conventional planning (without embedded hot spots), so long as the dose gradients outside of the tumor are acceptable (close or equal to the conventional plan). In this report, we have compared several approaches that can achieve this goal.
This planning study involved a simplified virtual phantom, targets, and OARs to compare optimization approaches in terms of the resultant dose distributions for varying target sizes. The impact of patient heterogeneity, abnormal target shapes, and overlap between PTV and OARs will lead to differences in the maximum achievable doses from those found in this study. Furthermore, we limited our investigation to a 6 MV beam with a flattening filter, commonly used for VMAT. Flattening filter free (FFF) beams may enable further increases in embedded target dose by providing a sharper penumbra and naturally peaked dose profile, as well as shorter delivery times for hypo-fractionated treatments [
30]. Other non-coplanar treatment configurations such as 4π SBRT [
31] or CyberKnife [
32] may also enable further increases in central tumor dose and improved gradients by reducing beam overlap outside of the target, but comparison of these approaches was beyond the scope of this study. Finally, the results of this study do not take any geometric uncertainty into consideration, pre-supposing that the target is entirely solid tumor, without motion, allowing and motivating increases in dose within the target. These approaches would not be appropriate for targets with critical structures interspersed within the target, or targets in regions with large setup uncertainties or internal motion.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.