Prescription to 50-75% isodose line may be optimum for linear accelerator based radiosurgery of cranial lesions

Abstract

Purpose

The prescribed percentage-isodose-line (PIDL) in linac-based SRS varies among institutions. For plans with similar tumor coverage and conformity index, the one with sharper dose falloff outside the tumor volume would be preferred because the probability of brain necrosis is related to the irradiated volume (for example V_12Gy) outside the tumor. The aim of this study is to investigate the optimal isodose line that yields the steepest dose falloff for linac-based SRS using dynamic conformal arc technique (DCA).

Methods

30 patients with brain tumors were retrospectively studied. The MLC-based DCA was used for planning. For each patient, 5-7 plans with different PIDLs but similar conformity indices were generated. All plans were normalized such that 95% of target volume receives the prescription dose (PD). Gradient index was calculated. The plan with minimum GI was considered optimal.

Results

Optimal GI decreases (3.9 to 2.2) as target volumes increases (0.15 to 50.1cm³), and the optimal PIDL shifts to higher percentile. Median optimal PIDL is 40.0±7.2% (range 33.2-53.1%) for targets <1cm³ and 62.3±8.3% (range 44.6-72.9%) for those >1cm³. The average planned PIDL used for treatment was 83.6±3.3%. The lower optimal PIDL results in smaller V_0.5PD and higher mean dose to the tumor.

Conclusion

The optimal PIDL appears to be between 50% and 75% which is lower than the commonly used PIDLs in published studies. Larger targets tend to have higher optimal PIDLs. By choosing an optimal PIDL, we could reduce the volume of irradiated normal brain while delivering higher mean dose to the tumor.

1. INTRODUCTION

The goal of radiotherapy is to achieve tumor control and minimize the damage to normal tissue. With modern treatment planning and delivery systems, the prescription isodose line can be made very conformal to the tumor volume, which allows for optimal sparing of critical organs-at-risk [1,2]. Fractionation is a method to allow normal tissue to repair while accumulating dose to the tumor [3], so a radiotherapy prescription dose (PD) is typically delivered in multiple fractions. For brain metastasis, fractionated whole brain radiation therapy (WBRT) has been a standard treatment. Stereotactic radiosurgery (SRS) has also been utilized for many decades. A recent clinical trial showed that the addition of SRS to WBRT improves survival and functional autonomy for patients with one to three brain metastases compared to whole brain radiation therapy only [4]. Studies also showed that there is no apparent survival benefit for patients with brain metastases when WBRT was combined with SRS compared to SRS alone [5-7]. SRS has become popular for patients with a limited number of brain metastases, primarily to reduce the potential late effects of WBRT.

In contrast to fractionated treatments, SRS delivers a single fraction of highly focused ablative dose to a relatively smaller tumor volume and can be used repeatedly to remote new metastases. Thus the protection of normal brain tissue must be achieved by good dose conformity, a steep dose gradient outside the tumor volume and precise localization [8-12]. For SRS, conformity is often evaluated as part of the planning process and often reported in planning studies [13-15]. The dose gradient, which quantifies the dose falloff, is another important metric, because the adverse effect from SRS is related to the volume of the irradiated normal brain tissue. For example, reducing the 10-12 Gy volume for SRS can reduce treatment related toxicity [16-18].

For the same delivery system and delivery technique, the dose gradient can change when prescribing to a different percent isodose line. For Gamma Knife based brain SRS, 50% is usually chosen as the prescription (marginal) dose [19-21], and an even lower isodose line is suggested [22]. Linacs have also been widely adopted for SRS procedures for cranial lesions. However, the PIDL varies among different institutions [13,23,24]. Recently, Ohtakara, K. et al compared three discrete PIDLs and found 70% to be dosimetrically superior to 80% and 90% [25]. However, they didn’t examine the other PIDLs which could have potentially been superior according to their criteria. In this work, we studied a range of PIDLs and attempted to explore the optimal PIDL for linac-based intracranial SRS that provides the steepest dose fall off outside the tumor volume. We also reported the optimal block margin and the relationship between optimal PIDL and tumor volume.

2. METHODS AND MATERIALS

2.1 Planning Technique and Objectives

Thirty patients who received SRS treatment using dynamic conformal arc technique from 2011 to 2013 were retrospectively selected for this study. All patients were immobilized in the BrainLab mask. Treatment planning CTs were acquired with 2 mm slice thickness on a Philips Brilliance Big Bore CT scanner (Philips HealthCare, Best, Netherlands). The MR images were registered with planning CT images in the iPlan system (Version 4.1.2, BrainLAB AG, Feldkirchen, Germany) to facilitate target volume delineation. The planning target volume (PTV) was equal to gross target volume (GTV), which was defined on T1 weighted post-contrast MR images. The tumor volume sizes ranged from 0.15 to about 50.1 cm³ with a median value of 4.8 cm³. The original clinical plans were created on a Novalis TX linear accelerator (BrainLAB AG, Feldkirchen, Germany; Varian, Palo Alto, CA) equipped with 6MV SRS mode, HD120 multi-leaf collimator (MLC) and robotic couch with 6 degrees of freedom.

The number, the length and the separation of arcs all affect the dose fall-off. In this study, we chose the arc arrangement so that the dose falloff was in the optimal range based on our experience and the published study [26]. The treatment plans typically consisted of 4 to 5 non-coplanar arcs. Typical couch angle separation used was 30-40 degrees with no opposing arcs. Each arc usually spanned 120-140 degrees. For the majority of cases, the collimator angle was set at zero degrees for more efficient treatment delivery. Because the dose falloff from a target depends on the MLC block margin for a defined dose volume coverage, plans were recreated with varying MLC margin ranging from 2 to 3 mm for each case. Depending on the size of the tumor volume, the minimum MLC margin ranged from 3.0 to 0.5mm. As a result, 5-7 plans were generated for each patient. Every plan was normalized so that maximum dose was at 100% and 95% of tumor volume received prescription dose.

2.2 Evaluation Methods

We define the prescription isodose line, or PIDL, as the ratio of prescription dose and maximum dose, which is similar to other studies and Radiation Therapy Oncology Group (RTOG) protocols [13,25,27-28]. To quantify the dose falloff, gradient index (GI) was used [22]. It is defined as follows:

where V_0.5PD is the volume that receives 50% of the prescription dose and V_PD is the volume receiving the prescription dose. A plan with lower GI has lowerV_0.5PD, or sharper average dose fall off from 100% to 50% isodose line.

For each case, plans with varying amount of MLC block margin were created. The GI and PIDL that covered 95% tumor volume were calculated for each plan and the one with lowest GI was identified as optimal. The conformity index (CI) was also calculated using definitions defined by both RTOG and Paddick et al [29,30]. For completeness, the RTOG definition is: CI = V_PD/TV while that of Paddick is PCI = TV_PD²/(TV*V_PD), where TV is the target volume, TV_PD is the target volume that is covered by the prescription isodose. CIs were not included in determining the optimal PIDL in this study because CIs are very similar between plans of each case and it generally does not affect the dose falloff. For each case, we optimized the dose gradient while keeping similar conformity indices and target coverage.

The original treatment plans used for patient treatment did not always have 95% of target volume covered by the prescription dose line. To make a fair and meaningful comparison between the optimal PIDL and planned PIDL, the equivalent planned PIDL was calculated as the ratio of the dose covering 95% of the target volume and the maximum dose.

3. RESULTS

Dosimetric parameters are summarized in Table 1. Because there are 5-7 plans for each case, mean conformity indices and the standard deviations are presented to show that each group of plans maintains similar conformity except a few outliers. The MLC block margin and gradient index shown in the table are from the plans that provided the optimal PIDL for each case.

Table 1.

Plan parameter summary of the 30 cases. PIDL stands for prescription isodose line; GI is gradient index [22]; RTOG CI is conformity index as defined by RTOG [29]; PCI is conformity index as defined by Paddick et al [30]; Planned PIDL is calculated as if the clinically treated plan has the same target coverage of 95%.

Case #	Target Volume (cm3)	Optimal PIDL (%)	MLC Block Margin (mm)	GI	RTOG CI (mean±SD)	PCI (mean±SD)	Planned PIDL (%)
1	0.15	52.4	-0.5	3.87	1.59±0.15	0.57±0.05	79.3
2	0.21	47.6	-1.0	3.50	1.30±0.06	0.69±0.03	80.3
3	0.37	42.6	-1.0	3.14	1.61±0.11	0.56±0.04	78.6
4	0.44	34.6	-2.0	3.07	1.27±0.03	0.71±0.01	82.0
5	0.48	46.8	-1.0	3.19	1.30±0.05	0.70±0.02	82.8
6	0.53	53.1	-0.5	3.06	1.32±0.09	0.69±0.05	74.2
7	0.61	33.2	-2.0	3.00	1.23±0.05	0.74±0.03	82.1
8	0.75	39.0	-1.5	2.90	1.24±0.09	0.73±0.05	82.0
9	1.3	52.4	-1.0	2.75	1.21±0.07	0.74±0.04	83.6
10	1.8	47.6	-1.0	2.76	1.33±0.12	0.68±0.06	81.0
11	2.1	44.6	-1.5	2.62	1.25±0.09	0.72±0.05	83.9
12	2.6	52.1	-1.0	2.70	1.28±0.09	0.71±0.05	82.3
13	3.1	58.6	-1.5	2.57	1.14±0.05	0.79±0.03	86.6
14	4.2	60.9	-1.5	2.51	1.08±0.01	0.83±0.01	88.8
15	4.7	57.4	-1.0	2.48	1.20±0.06	0.76±0.04	82.5
16	4.8	57.2	-1.0	2.55	1.22±0.10	0.75±0.06	83.7
17	6.1	59.8	-1.0	2.41	1.15±0.05	0.79±0.03	83.8
18	6.9	56.2	-2.0	2.47	1.15±0.06	0.79±0.04	85.6
19	7.3	64.7	-1.5	2.48	1.16±0.07	0.78±0.04	88.8
20	7.8	62.0	-1.0	2.45	1.16±0.06	0.78±0.04	82.8
21	9.5	66.2	-1.5	2.68	1.22±0.06	0.74±0.03	85.1
22	11.4	66.0	-1.0	2.39	1.05±0.05	0.86±0.04	82.6
23	12.6	67.7	-1.5	2.44	1.11±0.05	0.82±0.04	88.2
24	14.1	72.1	-0.5	2.39	1.06±0.04	0.85±0.03	85.6
25	18.8	69.2	-1.5	2.36	1.12±0.10	0.81±0.06	84.8
26	21.3	72.3	-0.5	2.42	1.14±0.06	0.80±0.04	91.1
27	27.3	66.6	-1.0	2.13	1.27±0.14	0.72±0.07	85.8
28	34.4	71.5	-1.5	2.31	1.07±0.08	0.85±0.06	84.5
29	41.7	72.1	-1.5	2.29	1.06±0.08	0.86±0.06	84.0
30	50.1	72.9	-1.5	2.24	1.07±0.09	0.85±0.07	82.8

By changing the MLC margin and keeping the same target coverage, the dose inhomogeneity changes dramatically inside the target. As seen in Figure 1, a larger positive margin yields a more homogeneous dose distribution, however, it also gives more dose to the surrounding tissue. As the margin decreases, the dose distribution inside target becomes more heterogeneous (hotter). However, by examining the tissue DVHs, we can tell the dose to normal tissue decreases as the margin decreases until it reaches a minimum. This result is further shown in Figure 2 by plotting the V_0.5PD as a function of MLC block margin. V_0.5PD decreases as the MLC margin decreases until it reaches the minimum, from which point V_0.5PD increases as the MLC margin shrinks further. Thus there is an optimal MLC margin where the minimum GI is achieved. For the 30 cases in this study, the mean optimal MLC margin is 1.2 mm with a standard deviation of 0.4 mm.

Figure 1.

Typical DVHs of target and 1cm of tissue surrounding target. Total of 6 plans were generated for this case. From higher PIDL to lower one, the MLC margin used, in mm, is 2, 1, 0, 0.5, 1, and 2, respectively. Solid line is the plan that yields the lowest GI which is considered the optimal PIDL in this case.

Figure 2.

The GI and V_0.5PD vs PIDL for the same case as in Figure 1. For data points from left to right, MLC margin in mm is 2, 1, 0.5, 0, 1 and 2, respectively.

Figure 3 shows the GI and the associated PIDL for a subset of 10 cases, which is illustrative of the trends seen in all 30 cases with different tumor volumes. Except for the 5 smaller targets (average volume 0.35±0.18 cm³), each curve has a minimal GI that corresponds to the steepest dose falloff. In general, larger tumors have lower GIs and the optimal GI occurs at a higher PIDL. As target volumes increases from 0.15 to 50.1 cm³, the optimal GI decreases from 3.9 to 2.2. There is a wide and relative flat region where GI is close to its minimum for target volumes greater than 2 cm³.

Based on gradient index alone, the optimal PIDL appears to be between 40% and 75% for most cases and it depends on the size of the tumor volume as shown in Figure 4 (a). There is a strong correlation between optimal PIDL and target volume as shown in Figure 4 (b) (linear regression R²=0.75). Median optimal PIDL is 40.0±7.2% for target volumes less than 1 cm³ and 62.3±8.3% for those larger than 1 cm³. For volumes smaller than 1 cm³, a larger variation in the optimal PIDL (33%-53%) was observed.

Figure 4.

(a) Comparison of planned PIDL and optimal PIDL; (b) Relation between optimal PIDL and volume on a semi-logarithmic plot. Linear regression lines are shown. The determination of coefficient R2 = 0.75 for all data, and 0.83 for targets larger than 1 cm3.

The optimal PIDL is found to be much lower than our planned PIDL and those reported in other studies [13,16,24]. The average optimal PIDL is 57.3±11.6% (range 33.2-72.9%, however the average planned PIDL is 83.6±3.3% (range 74.2-91.1%). The lower optimal PIDL corresponds to smaller V_0.5PD. For example, about 19 cm³ reduction in V_0.5PD was observed in the case that had the largest tumor volume. A lower PIDL also yields higher mean dose to the tumor volumes.

4. DISCUSSIONS

Many studies have reported that the volume of normal brain tissue receiving 8-12 Gy during radiosurgery is an important predictor of brain necrosis [16-18]. A dose level of 8-12 Gy corresponds to about 50% isodose line for typical prescriptions of 16-24 Gy. Thus we adopted the GI (V_0.5PD/ V_PD) as a clinically relevant measure of dose falloff. By finding the lowest GI and prescribing to the corresponding PIDL, the V_0.5PD is reduced, which could potentially reduce the probability of brain necrosis. If another dose level instead of 0.5PD dose is deemed clinically relevant, one can apply the same approach to evaluate the dose fall-off and obtain the optimal PIDL. On the other hand, because the optimal PIDLs were all lower than the planned PIDLs in general [13,16,23,24], the mean dose to the tumor volume would be higher if the optimal PIDLs were delivered, which could potentially increase the tumor local control as shown in a modeling study [31].

As the target volume gets smaller, the optimal PIDL becomes lower. It is not advisable to choose a PIDL less than 50% because of large heterogeneity in the dose distribution [29]. In fact, there is a large variation in the optimal PIDL (33%-53%) for targets smaller than 1 cm³. One reason is, for very small targets, the GI is very sensitive to the shape, the location of the target and the beam settings. Another cause could be the limitations of the treatment planning system in computing dose for very small fields. For very small targets, there could be situations, where the field aperture becomes smaller than the smallest beam model input data of the planning system as the block margin is reduced. Thus, the results for very small targets may not be accurate. However, choosing a sub-optimal PIDL will not change V_0.5PD much for small targets. For example, for the case with a 0.21 cm³ target shown in Figure 3, the optimal PIDL appears to be less than 50%. However, if one chooses 70% instead, the increase in V_0.5PD is less than 0.3 cm³. Even for the case of 2.1 cm³ (optimal PIDL 44.6%), the increase in V_0.5PD is less than 0.5 cm³ if the chosen prescription line is 65-70%. Thus we recommend 65-70% for small targets to avoid large dose heterogeneity. For targets with larger volumes, an individualized PIDL can be applied. From this study, if one prefers a standard PIDL versus an individually optimized value, 70% appears to be a good compromise amongst GI, V_0.5PD and dose heterogeneity, which is consistent with the other studies [25,32].

Figure 3.

Gradient index as a function of PIDL for 10 representative cases. Data labels are tumor volumes in cm3.

Excluding very small target volumes, the optimal PIDL for linac-based intracranial SRS is found to be in the range of 50-75%. In comparison, 50% is commonly prescribed to for Gamma Knife based SRS. The 40% PIDL was reported to be optimal using a similar GI analysis (mean target volume 2.9 cm³, range 0.2-12.9 cm³) [22]. For CyberKnife, the typical PIDL is 80%, but there is no indication that 80% is optimal [33-35]. One study reported the optimal PIDL for CyberKnife is approximately 40%. However, this study was based on phantom studies for extra-cranial treatment [32]. For linac-based SRS using a dynamic conformal arc technique, many institutions used PIDLs of 80-90% with a few clinicians using 70-80% [13,16,24,25]. Ohtakara, K. et al compared three commonly used PIDLs and found 70% to be better than 80% or 90% [25]. In their study, the target volumes ranged from 7.4 to 25.9 cm³. This is consistent with our finding of 69±3% for tumors larger than 7 cm³. However, the optimal PIDL is tumor size dependent. As the tumor volume becomes smaller, the optimal PIDL decreases. For tumor volumes between 1 and 7 cm³, the optimal PIDL is found to be 55±5% which is much lower than 70%.

A single index GI was used to determine the optimal PIDL for a pre-defined target coverage. Our assumption is that the minimum GI corresponds to an optimal PIDL. All plans created had similar plan quality in terms of tumor coverage and conformity index, but different dose heterogeneity. The dose heterogeneity within the target volume was assumed to be acceptable [36,37]. In fact, the pattern of dose heterogeneity usually showed higher dose in the central region of the target, and relatively lower dose in the peripheral region. Such dose heterogeneity is often preferred to uniform dose distributions within the target, because tumor burden is usually higher in the central tumor region.

In this study, 30 cases with dynamic conformal arc technique were selected and analyzed. Target volumes that require intensity modulated inverse planning were excluded because the value of V_0.5PD depends highly on the optimization parameters and the treatment planning technique. In such plans, the dose falloff is sensitive to specific tumor shape and proximity of organs-at-risk. Thus it is difficult to correlate GI with the optimal PIDL. In contrast, a dynamic conformal arc plan typically employs five arcs that spread out at different couch angles, and an isotropic dose falloff is normally achieved around the target volume. The GI is limited by the physical characteristics of the linear accelerator such as the width of the MLC leaves, the beam quality and penumbra, the intra- and inter-leaf transmission, etc. Adding more arcs than five only improves GI marginally as reported elsewhere [26]. Therefore we typically use five arcs.

For most cases, the MLC margin that yields the minimum GI is between 1.5 and 0.5 mm which is consistent with findings from other studies [32,37]. By changing from a positive to a negative margin, more monitor units (MUs) are required to achieve the same target coverage. While the dose drop off is faster from 100% to 50% isodose lines when shrinking the margin, the lower isodose line volume may increase as a result of increasing MUs. Although it is not known if the lower dose volume size correlates to brain necrosis, clinicians need to be aware of this and its potential risk. The dose gradient might also depend on the depth of the tumor location due to changes in scatter and the beam energy spectrum. Furthermore, more MUs introduce more leakage and scatter which increases the whole body dose. Thus cautions should be taken for patients that have longer life expectancy and measurements are recommended to evaluate the scatter and leakage exposures. Additionally, the accuracy of the dose calculation can be affected by the accuracy of the MLC leaf edge model in the planning system, which might impact the calculated gradient index. In this regard, it is important that the commissioning measurements include MLC-defined field shapes to capture the leakage and scattering effects of the MLC.

5. CONCLUSIONS

The optimal PIDL appears to be between 50% and 75% for the cases evaluated, which is much lower than planned PIDLs in our current clinical practice (approximately 84%) and most reported studies. Larger target volumes tend to have a higher optimal PIDL. By choosing an optimal PIDL, we can potentially reduce the toxicity to normal brain tissue while delivering a higher dose to the target volume. This conclusion is based on the assumption that the minimal GI corresponds to the optimal PIDL. A prospective study is needed to correlate the optimal PIDL with observed clinical outcomes.