Background
Evidence continues to support the safety and efficacy of stereotactic radiosurgery (SRS) [
1,
2] and emerging techniques such as hypo-fractionated radiotherapy (HF-RT) [
3‐
5]. For SRS, recent clinical evidence supports the safety in treating up to ten targets in a single fraction [
1]. For HF-RT, radiation-induced harm to normal tissue, in particular for large or recurrent disease, is mitigated whilst delivering a clinically effective dose to the target [
4‐
7]. In conjunction with modern treatment planning systems, linear accelerators equipped with high-definition multileaf collimators (MLCs), image-guidance and robotic couches, both SRS and HF-RT are increasing in utility.
Cases are becoming more challenging with centres treating large number of targets, large volumes, recurrent disease, complex shapes, etc. The well-known safety limit of V12 < 10 cm
3 for SRS [
8‐
10] may not necessarily apply or be feasible for multiple targets, re-treatment scenarios, or HF-RT. The focus in the literature has deservedly been on investigating what technique offers the best rapid dose fall-off: the more rapid the better [
11‐
13]. However, every paper reports their endpoints in a different way, making inter-institutional comparisons challenging if not impossible. Other approaches involve developing predictive models of dose fall-off [
14‐
16]. The work by Shiraishi et al. develops and validates a knowledge-based approach of an accurate model, however the model itself is hidden to the reader [
14]. The approach by Bohoudi et al., on the other hand is tailored specifically towards V12 [
16]. Furthermore, there are limited published data specific to HF-RT. In one paper by Ruschin et al., irradiated volume is reported as a function of target volume but only for specific dose-fractionation schemes and only for Brain Minus CTV volume, which can be different amongst institutions who apply different PTV margins than those used [
17].
In the present study, we propose a framework that all clinics can easily access to evaluate irradiated volume for arbitrary isodose levels; for both untreated brain (Brain Minus PTV) and total tissue; for varying target numbers and shapes. Although irradiated volume depends on many factors such as machine characteristics and technique, we sought to provide a framework through which our institution and others can readily compare population data to each other. The framework can also be applied to prospective individual cases, in which some guidance as to “what is achievable” for that case could be of assistance. Internally, such a framework could provide the basis for treatment plan quality assurance (QA), especially for challenging or off-protocol cases. Additionally, investigating population trends over time can lead to more consistent and higher quality treatment plans. The intention is not to precisely predict irradiated volume for all institutions and techniques, but rather to provide the framework through which irradiate volume can be compared across institutions.
Discussion
We have developed and tested a readily accessible framework for irradiated volume in HF-RT. The framework describes irradiated volume of the external contour (EXT) and Brain-Minus-PTV (BMP) for any given target volume and isodose level that any institution can implement for their technique and patient population. The purpose is not to seek out which technique is “better” or “worse”, but to simply facilitate a means for technique comparison.
The primary gain of the proposed framework is that despite the complex underlying physics behind it, irradiated volume can be reduced to a simple set of variables, which any institution can measure for their HF-RT patient population. Although we have limited the variables to target size, number, and shape for a given technique one could easily expand upon the model by adding other refining features, including MLC size, MLC margin around the target and number of arcs or isocenters. The clinical benefits of such a system include: (1) a framework through which institutions can perform treatment plan quality assurance, especially for complex situations such as HF-RT for which guidelines are lacking; (2) a framework through which treatment plans can be compared across institutions, which is often hampered by differing definitions of quality metrics. There are numerous published papers regarding the dosimetric benefit of one treatment technique over another, but the conclusions are often mixed, with some papers claiming little difference [
12] and others citing larger differences [
11]. The proposed framework may provide a means through which irradiated volume is parameterized and compared across institutions in a fair way. (3) a framework through which institutions can
improve upon plan quality by reducing the variation in their own measured model over time. In our clinic, we have initiated using the proposed framework during the treatment plan review process by comparing irradiated volume parameters for upcoming plans against the model-predicted values. If a plan exceeds the prediction interval of the model, an investigation is undertaken to determine whether plan improvement can be made, and over time the spread of dosimetric data will diminish. Note that the prediction interval used in the present study was taken to be 95%, which was selected to cover an accurate range of dosimetric outcomes, at the cost of having wider intervals than using a less-predictive model such as with an 80% prediction interval.
As an example, a centre relatively new to HF-RT wishes to treat an inoperable and relatively large intact brain metastasis that has a volume (after PTV expansion) of 60cm
3 with a fractionation scheme of 30Gy in five fractions, for which there is limited data on dose-volume constraints for normal tissue. The clinic is interested in the volume of tissue receiving >21Gy in five fractions, which for a 30Gy prescription is 70% of the prescribe dose, i.e. V70. From Fig.
1(c) and (d), if using a SAHO technique one may expect a range of V70 to be 119 cm
3 ± 9 cm
3 or 39 cm
3 ± 11 cm
3 for EXT and BMP respectively. If the clinic were willing to accept a hotter plan, with a target Dmax of up to 150% and planned with the FAHE technique, then the corresponding expected V70 would be lower than for SAHO (see Fig.
3d), in this case 83 cm
3 ± 6 cm
3 and 27 cm
3 ± 8 cm
3. If the target was irregularly shaped, or an additional target was also present, then the factors F
N or R could be appropriately applied. For any given isodose the same process can be applied, using the curves in Fig.
2 to extract the relevant parameters needed to asses irradiated volume. These ranges of irradiated volume serve as a guide of what may achievable: if a higher irradiated volume is achieved than the higher bound of the predicted range, then perhaps it may prompt a closer look at the complexity of the case or whether there can be improvements, whereas if a lower irradiated volume is achieved then that technique used is capable of achieving lower irradiated volumes, which may be a benefit.
An advantage of the present manuscript is that irradiated volume is described independently from prescription dose and PTV margin, and for both EXT and BMP, which facilitates comparison to published data. Certain treatment planning systems, such as Leksell GammaPlan, report irradiated volume to EXT whereas other planning systems can generate brain contours, and often subtract the PTV to define untreated brain, which we report as BMP. Irradiated BMP volumes are always less than EXT volumes (see Fig.
1), since the target and all dose spilling outside the brain contour are excluded from BMP. There is also more variation in the BMP curves: depending on the target location, more-or-less dose will spill beyond the brain into the skull and surrounding tissue. Our proposed framework also includes measuring EXT and BMP at multiple isodose levels such that any future isodose level can be interpolated, by presenting the fit parameters similarly as shown in Fig.
2.
Although we have demonstrated feasibility of our framework by comparing to internal and external data, there are certain limitations of the framework that need addressing. Firstly, our underlying clinical data is for single-arc HF-RT treatments. Although we re-planned 30 cases with a published FAHE technique, we do not use that approach clinically. Furthermore, our investigation of factors F
N and R may be limited to our treatment technique and patient population, which by virtue of being HF-RT, consisted of larger targets than those used for SRS. For example, our finding that having >3 targets resulted in higher (range: 1.4–2.9×) irradiated volumes than single-target cases may be due to increased dose interplay between targets resulting from the larger targets in our series (median and maximum volume = 14.1 and 84.6 cm
3 respectively) than for typical SRS series. By contrast, the single metastasis model proposed by Bohoudi et al. consisted of target volumes in the range of 1 to 20 cm
3 (median = 7 cm
3) and was validated against multiple targets [
16]. Naturally, there are many factors affecting the extent of irradiated volume, including penumbra, MLC margin around the target, beam modulation and number of arcs. However, despite the limitations, reasonable agreement was found comparing our FAHE model to data from other institutions demonstrating feasibility for inter-institutional comparisons [
12,
20,
21]. This shows that at a minimum the specific factors we found can be used as a starting point going forward, and refinements can be easily incorporated. It would be interesting to present irradiated volume data acquired from other systems, such as the Gamma Knife or Cyberknife, in a similar format to that in the present manuscript.
It is important to emphasize that there are other factors to consider than irradiated volume when evaluating treatment planning techniques. For example, the conformity of the prescription isodose line to the target can vary between techniques. Another factor to consider is dose homogeneity. For HF-RT, our institution has consistently employed a 2 mm PTV margin, which has limited the comfort of the treating physicians to allow hotspots in the target in excess of 120–130%, as such steep gradients may cause normal-tissue necrosis within the margin itself. The dose limit of 130% is different from the situation of single-fraction SRS, in which the hotspot may easily extend up to 150–160%. Furthermore, the number of arcs and the degree of modulation can affect dose fall-off. The treating centre should determine what maximum allowed dose and degree of modulation it is comfortable with. Other important plan quality metrics for dose fall-off include the Paddick Gradient Index (GI), which is the ratio of volume receiving half of the prescribed dose to the volume receiving the prescribed dose. While GI is easy to measure, the expectation of the present work is that that normal tissue tolerances for HF-RT will ultimately be derived from clinical studies and reported in absolute volumes, much like the safety limit of V12 < 10 cm3 is used for SRS. One of the major strengths of the present work is there are no assumptions made as to what an “optimal” plan is for any given case, since treatment plan optimality depends on a multitude of factors. Rather than attempt optimality, our approach is heuristic in nature, and uses plans delivered to patients that the practising physicians (many of who have greater than 10 years experience in the field) deemed clinically acceptable.
Among other study limitations is the lack of clinical outcomes associated with the reported irradiated volume data. Although outcomes modeling is outside the scope of this work inclusion would clearly strengthen impact, and this work is ongoing. Another study limitation is that relatively few SAHO cases had more than 2 or 3 targets; having more multi-target cases would increase the accuracy of the determined F
N. Although the framework can be extended to isodose lines <50%, 50% represents a clinically relevant lower limit [
12]. Furthermore, in choosing to make the present manuscript readily comparable to other papers in the literature, we assumed that relative isodose lines are maintained irrespective of the absolute prescription dose whereas the planning process must incorporate machine constraints such as gantry speed.