4.1 Quality of plans for MI and SI approaches

In terms of target conformity and doses to OARs, both MI and SI approaches produced clinically acceptable SRT plans for patients with multiple brain lesions. These results were in line with previous studies3, 15 that reported that the SI approach produced comparable target dose coverage and normal tissue doses when compared with the MI approach in SRT of multiple brain metastases. Algan et al. also demonstrated that there was an added advantage of SI plans that was a 35% reduction in beam-on time.15

Although both approaches could produce plans satisfying the plan requirements, we found that the SI plans would result in relatively lower PTV dose conformity (CI of 0.83 vs. 0.84 but not statistically significant) and higher maximum doses to some OARs (e.g., left optic nerve and chiasm; significantly higher doses) than the MI plans. The relatively poorer target conformity in SI plans could be explained by the fact that better target conformity could be achieved with a greater number of isocenters and radiation beams40 as in the case of the MI plans. While one isocenter was assigned to each target in the MI plans in which the dose distribution for each target could be adjusted independently, for the SI plans, the isocenter was placed at roughly midway between targets, and the planning was performed by considering all targets together. These inevitably limited the flexibility of manipulating the treatment parameters of individual targets and therefore would lead to a less ideal target dose distribution in SI plans. Moreover, Morrison et al. reported that target conformity and gradient indices worsen with increasing distance of the isocenter from the PTVs.3 The brain lesions were usually sparsely located. The location where the isocenter was placed in SI plans may be within the normal brain tissue, when compared with isocenters in MI plans which were placed inside each brain lesion, or even at the center of each brain mass. Planning with a SI for multiple brain lesions usually encounters disadvantages brought by a larger distance of isocenter with PTVs.

A possible explanation for relatively higher doses received by a few OARs in SI plans than in MI plans is that larger collimator size and the use of wider MLCs were usually required in the SI plans to cover all targets in linac-based SRT. This would reduce the ability to shape the dose around the target and at the same time avoid the dose to different OARs. Besides, it would also lead to more leakage dose between MLCs and greater scattered radiation,41 hence subsequently increased the doses to OARs. An island blocking problem would occur, when multiple targets (≥2) share the same pair of MLC, causing an area of non-target tissue that is not covered by the MLCs.14

Nevertheless, since the overall dosimetric differences between MI and SI approaches were relatively subtle, most researchers advocated that the effect was clinically insignificant.3, 40 This has been supported by several clinical studies in which the local control and toxicities were comparable between these two approaches.1, 12, 42 Lau et al. only deduced that minor improvements in plan quality can be attained by MI.1 In addition to the advantage of the resulting shorter treatment time, the SI approach in the treatment of multiple brain lesions was generally appreciated by oncology departments.

4.2 Effects of isocenter shifts on treatment plans

After the introduction of isocenter shifts that aimed to simulate the intra-fractional setup discrepancies in the daily clinical situation, the PTV doses were all affected in both MI and SI plans. The result in this study illustrated larger effects with translational shifts than with rotational shifts according to the magnitude of δCI in both plans. This echoed the report from Wang et al. who studied the dosimetric results in spinal stereotactic body radiotherapy and addressed that a 2-mm translational error could result in > 5% tumor coverage loss and > 25% maximal dose increase to OARs.27 The lower dosimetric impact in the rotational shifts could be due to the relatively small tumor volumes in SRT and the geometrical relationship between the isocenter(s) and PTVs. Extreme cases were found after ±1.5 mm translational shifts, where δCIs exceeded 0.25, and ±2 mm translational shifts, where δCIs were greater than 0.33 (Table 4), and the percentage of plans having VoR of ≥10% was greater than 91% in both approaches (Table 6). This revealed that large magnitude of translational shifts degraded the dose coverage and conformity to PTV and would result in non-clinically acceptable plans. In contrast, there would be less concern in deterioration in PTV doses for rotational shifts within ±2° since all δCIs were ≤ 0.16 for the SI approach or even < 0.1 in MI plans (Table 5).

With respect to the comparison of the impact of the isocenter shift on PTV dose between the MI and SI approaches, no statistically significant differences in δCI were found for all extents of translational shifts in the study. It is suggested that the geometrical relationship of the shifted-dose distribution and isocenter were moved in the same way for both MI and SI planning. Translational shifts do not appear to affect MI and SI plans differently. A larger extent of shifts resulted in further loss in CI when compared with its original plan (without shift) but affected SI and MI plans similarly.

Yet regarding the rotational shifts, SI plans were in general relatively more vulnerable than the MI plans as significant differences in δCIs were found after all magnitudes of shifts. This was reflected in the values of δCIs and analysis of VoRs, especially for 5%–10% and ≥10% of VoRs when shifts ≥ 1.5. The main reason for this lies in the difference in the PTV-isocenter relationship.43 The isocenter was the center of rotation where it was placed at the center of the PTVs in MI plans, whereas the isocenter for SI plans was distant from the PTVs in view of covering more than one target. Any shift would bring greater dose changes in PTVs than the MI plans. The effect would be magnified for PTVs situated further away from the isocenter. This observation can be further illustrated in the scattered plots (Figure 5a,b) demonstrating the relationship between the distance of isocenter from PTVs and the relative degree of change of target dose conformity (δCI) in SI plans when the rotational shifts were +2.0 and –2 respectively. PTV-isocenter distances were calculated using the root mean square of the differences in LR (lateral; x), AP (vertical; y), and SI (longitudinal; z) directions between the target PTV and its respective isocenter location, specifically as (2) Scattered plots showing the relationship of δCI against PTV-isocenter distance in SI plans after application of rotational shifts of (a) +2 and (b) –2

The graphs showed generally a pattern of decreasing robustness of SI plans (larger δCI) to an enormous extent of rotations as the PTV-isocenter distance increased. Their regression lines indicated a recommended threshold PTV-isocenter distance of 3.6–3.7 cm for rotational shifts of 2.0 and 2, assuming a maximum allowable δCI as 0.2. Owing to this phenomenon, the differences between the two approaches were small when the rotational shifts were small, but the discrepancies increased when the shift was amplified. Gevaert et al.44 and Huang et al.41 also summarized that a small angular error could result in considerable dosimetric degradation particularly for small targets at a distance from the treatment isocenter using the SI approach. To minimize the risk of compromised percentage target coverage in case of experiencing a large intra-fractional error in multiple-target SRT using SI approach, Roper et al. recommended to locate the isocenter closer to the small PTV instead of placing it midway between the PTVs.28

The impact of translational and rotational isocenter shifts to the OAR doses was relatively mild with δDMax of different OARs substantially less than 1 Gy in both MI and SI plans. The largest δDMax were only revealed as 0.73 Gy and 0.88 Gy of brainstem doses in SI plans after translational shifts of +2 and –2 mm (Table 4), and this result would be expected to have limited clinical significance. However, respective doses to different OARs become noteworthy especially when targets are in proximity. This may result in non-planned irradiation dose and hence collateral damage to these adjacent structures. For a patient whose brainstem was close to PTV and already received treatment dose close to the dose tolerance of the brainstem, a translational isocenter shift of > 1.5 mm should be avoided especially for the SI approach so as not to further increase the hazard of a potential extra dose of around 1 Gy to the organ bringing its total dose to exceed the tolerance. Similarly, attention should be also put on other OARs that are adjacent to the PTV that the addition of 0.5 Gy resulting from extreme isocenter shift might lead to risk beyond its respective tolerance limit. Additionally, as the shifted-dose distributions were no longer conformed to the PTVs, it is logical to observe that the impact of isocenter shifts to OARs doses became greater with the increasing magnitude of shifts.

When comparing MI with SI plans, there were not many significant differences in OAR doses caused by isocenter shifts despite higher values of δDMax resulting in SI plans for the majority of OARs after all types of shifts. Only the differences in the left optic nerve and chiasm showed generally consistent significance after both translational ≥ ± 1 mm and rotational shifts of ≥ ± 1. OARs were usually situated at various locations relative to the PTVs and isocenter. It is believed that the shifts might just contribute to a random effect on the OAR doses,45 and there is no definite pattern that any of the treatment approaches would be favored. Nevertheless, an important point to note is that over the highly hypofractionated course of treatment irradiating brain lesions in SRS, the random errors may not be provided with an opportunity to be averaged out; these random errors, therefore, become more significant. The impact of overdose on OARs may be as crucial as errors that underdose a PTV28 and hence cannot be underestimated. The steep dose gradient in SRT might imply that more precautions are required to protect the OARs.27

All in all, the influence of the intra-fractional isocenter shifts to both MI and SI plans for SRT of multiple brain metastases could not be viewed as negligible. Problems might arise from loss in PTV coverage and OARs overdose (particularly brainstem) when isocenter shift exceeds translational 1.5 mm or rotational 1.5, with the SI approach being more prone to the impact of shifts when compared with the MI approach. Thus, although the SI approach can offer a shorter treatment time and acceptable dose distributions in SRT for multiple brain metastases, greater effort has to be made to minimize the intra-fractional errors. This may include the use of image guidance with online position tracking and correction46 or increased frequency of monitoring for radiation treatment. On the other hand, further researches on topics such as analysis and estimation of PTV margins to account for the errors and investigation on more frequently fractionated treatment or the adoption of SRT instead of single-fraction SRS in SI approach to overcome the relatively inferior robustness to shifts based on the rationale that fractionation helps reduce the impact of random errors may pave the way to the future development of the use of SI technique.