Safety and benefit of using a virtual bolus during treatment planning for breast cancer treated with arc therapy

Abstract Purpose This study evaluates the benefit of a virtual bolus method for volumetric modulated arc therapy (VMAT) plan optimization to compensate breast modifications that may occur during breast treatment. Methods Ten files were replanned with VMAT giving 50 Gy to the breast and 47 Gy to the nodes within 25 fractions. The planning process used a virtual bolus for the first optimization, then the monitors units were reoptimized without bolus, after fixing the segments shapes. Structures and treatment planning were exported on a second scanner (CT) performed during treatment as a consequence to modifications in patient's anatomy. The comparative end‐point was clinical target volume's coverage. The first analysis compared the VMAT plans made using the virtual bolus method (VB‐VMAT) to the plans without using it (NoVB‐VMAT) on the first simulation CT. Then, the same analysis was performed on the second CT. Finally, the level of degradation of target volume coverage between the two CT using VB‐VMAT was compared to results using a standard technique of forward‐planned multisegment technique (Tan‐IMRT). Results Using a virtual bolus for VMAT does not degrade dosimetric results on the first CT. No significant result in favor of the NoVB‐VMAT plans was noted. The VB‐VMAT method led to significant better dose distribution on a second CT with modified anatomies compared to NoVB‐VMAT. The clinical target volume's coverage by 95% (V95%) of the prescribed dose was 98.9% [96.1–99.6] on the second CT for VB‐VMAT compared to 92.6% [85.2–97.7] for NoVB‐VMAT (P = 0.0002). The degradation of the target volume coverage for VB‐VMAT is not worse than for Tan‐IMRT: the median differential of V95% between the two CT was 0.9% for VMAT and 0.7% for Tan‐IMRT (P = 1). Conclusion This study confirms the safety and benefit of using a virtual bolus during the VMAT planning process to compensate potential breast shape modifications.


| INTRODUCTION
Adjuvant radiotherapy (RT) for breast cancer is a standard treatment used to improve local tumor control and overall survival. [1][2][3][4] Volumetric modulated arc therapy (VMAT) has been evaluated for breast treatment in several publications as attested in a recent review. 5 For now, only six publications report a clinical experience dominated by simultaneous integrated boost studies [6][7][8][9] or accelerated partial breast irradiation 10 and only one in the setting of nodal involvement. 11 The location of the mammary gland leads to clinical target volumes (CTV) adjoined to the skin. This particularity of breast's target volume would generate a planning target volume (PTV) located partially outside the external body contour if isotropic margins were applied. In inverse planning optimization, this prevents from taking isotropic PTV margins. It may lead to target volume's lack of coverage in case of inter-and/or intrafraction movements. This issue can be taken into account using a skin flash method for fixed fields. However, for arc therapy techniques, other solutions should be found.
A method using virtual bolus to force the leaves to be positioned away from the external part of the breast has been described. 12 However, it was tested in a theoretical way as automatic expansions were used to mimic inter-and/or intrafraction modifications. In this dosimetric study, we use a nearby similar virtual bolus method to check its reliability in true life: We use real setup errors and breast's shape modifications of 10 real patients treated for left breasts and lymph nodes including the internal mammary chain (IMC). These patients had been reimaged with a second scanner (CT) because of observed interfractional modifications.
In order to validate the safety of using of the virtual bolus technique for the inversed optimization process, this paper first evaluates the consequences of using of a virtual bolus on the initial planning CT. Then, the treatment planning reproducibility is investigated by comparing the plans made with the virtual bolus method (VB-VMAT) to the plans without using it (NoVB-VMAT) on the second CT. The consequences of these modifications on the coverage of target volumes and dose to the organs at risk are evaluated.
Finally, the level of degradation of target volume coverage between the two CT using VB-VMAT is compared to results using our institutional standard technique of forward-planned multisegment technique (Tan-IMRT).

2.A | Patient selection, contouring and prescription
Ten planning studies were performed for patients consecutively treated for breast cancer in our department. Inclusion criteria were leftsided breast cancer, with a target volume including breast/chest wall, supraclavicular, axillary II and III nodes, and IMC in the first three interspaces, planned on the same version of Pinnacle ® . Eight patients were treated with breast conservative surgery and two with mastectomy. The same datasets and contours were used to do the standard plans (on Pinnacle ® v9.10) and VMAT plans (on RayStation ® v5.0). Dose calculations were made using a collapsed cone convolution algorithm on both treatment planning systems (TPS) with a grid size of 3.0 mm. Both the plans were optimized for an Elekta Synergy LINAC equipped with an Agility 160 multileaves collimator (MLC). All patients had a second CT during treatment because of unsatisfying portal images or clinical edema.
The CTV included the breast/chest wall (CTV-T), supraclavicular, axillary level II and III nodes, and the IMC in the first three intercostal spaces (CTV-N). By adding a 5 mm margin around the CTVs, we generated PTVs located partially outside the external body contour. We called those volumes PTV-T outside and PTV-N outside . Then, by limiting those volumes 5 mm inside the external contour, we created PTV-T and PTV-N. Furthermore, to evaluate the degradation of the target coverage on the second CT, a CTV-T evaluation , limited 5 mm inside the external contour, is also constructed (Fig. 1).
Prescribed doses were 50 Gy in 25 fractions for the breast/chest wall and 47 Gy in 25 fractions for the regional nodes (corresponding to 46 Gy in 23 f with α/β = 4). The irradiation of the tumor bed is not considered in this work. The organ at risk (OAR) (lungs, heart, left coronary artery (LCA), right breast, humeral head, thyroid, and esophagus) were also contoured according to the RTOG recommendations (http://rtog.org/). A structure called "skin," was constructed as the 5 mm fringe under the external contour inside the PTV.
Treatment planning and contouring from the first scanner (CT1) were exported on the second scanner (CT2) after registration. These patients had been reimaged during their treatment because of observed interfractional modifications. As described in a previous study, 13 we used a rigid registration between CT1 and CT2 by focusing on a cubic region including the treated breast. Contours were manually adjusted to the new anatomy and the plan from CT1 was recalculated on CT2 without further optimization.

2.B | Volumetric Modulated Arc Therapy
For the VMAT plans, we used two arcs starting from 300°to 170°c lockwise and inverse, with one control point every 4°. We exclusively used 6 MV photons. Collimator rotations of + 10°and −10°w ere used to increase the modulation possibilities.
Inverse planning was made in two steps. The first step of the optimization process was made with the virtual bolus in place (voxel's density of the virtual bolus was set to the density of water). In the second step, the virtual bolus was removed (no density was applied to the virtual bolus) and a new optimization was made without changing the shape of the segments, that is only to adjust the number of monitor units (MU) by control point (Fig. 2). The bolus construction, shown in Fig. 3, required two steps: first, 5 mm was added at the PTV outside . Then, a subtraction was made from the external boundary of the patient automatically generated by the TPS.
During the inverse planning optimization process, objectives were chosen in order to respect the prescription regarding the PTV-T and fulfill the predefined following clinical goals: D mean < 6 Gy for the heart; V20 < 30% and V30 < 20% for the left lung; and D mean < 3 Gy for contralateral organs (lung and breast). In addition, for healthy tissues, the maximum dose should be inferior to 55 Gy.
For planning target volumes, 95% of their volume should be covered by 95% of the prescribed dose (V95%). Prescription was made on the median dose of PTV-T (50 Gy) for both VB-VMAT (at the end of the second step of the optimization process) and NoVB-VMAT. However, for VB-VMAT, the dose was prescribed on the PTV outside for the first step of the optimization process. The initial optimization objectives were fixed among cases but they could then be adjusted to meet clinical goals. Clinical goals were in accordance with the external RT guidelines published in 2007. 14 The NoVB-VMAT plans were reoptimized based on VB-VMAT optimization parameters.

2.C | Institutional standard technique: tangential image-guided radiation therapy technique
Our institutional standard technique used two tangential fields (for the breast) and four additional static fields (for the nodes) as described in previous publication. 15 Tangential and node fields are constructed from the PTV outside with margins. An overlap of ≤7 mm at the skin between the tangential and node fields is accepted. 6 MV photons are used (or a mixture of 18 and 6 MV photons for large volumes). The IMC field is treated using a combination of photons and electrons.
Contralateral OAR are excluded from the primary fields.
The dose distribution to the breast is optimized using a field-infield technique consisting in suppressing overdoses regions by successive segments. The overdoses areas are hidden by 6% levels. The segment size was restricted to a 1.5 cm around the prescription point and a minimum of four MU per segment was required. Three or four segments are usually used; the main segment that corresponded to the whole tangential field consists of approximately 80% of the MU.

2.D | Comparison criteria and statistics
The main goal of this study was to evaluate the benefits when using a virtual bolus during the planning process to maintain the coverage of target volume when breast shape modifications or set up errors occur. Three steps were used for the demonstration: (a) First, the VB-VMAT plans were compared to the NoVB-VMAT plans. For the CTV-T and CTV-N, we have compared target coverage (V95%) as well as homogeneous and conformity index (HI and CI). As mentioned above, for evaluation purposes, the CTV-T was limited at 5 mm inside the external contour to exclude the first millimeters where the uncertainty on dose calculation is more elevated. The HI was calculated according to the following formula: HI = D2% − D98%/D where D2% is the dose to 2% of the volume, D98% is the dose to 98% of the volume, and D is the prescribed dose. Then, the conformity index (CI) for the combined CTVs was compared. The CI was automatically calculated on Raystation ® defined as follows: CI = TV PVI /V PVI . (TV PVI is the target volume covered by the prescription isodose and V PVI is the total prescription isodose volume.) Dosimetric results on the OAR were also detailed.
(b) The same analysis was performed on CT2 for the second step.
(c) The third step was the comparison of the coverage modifications between the two CT for VB-VMAT and Tan-IMRT. For that purpose, we used the V95% data for both techniques and their differential (ΔV95% (CT1-CT2)).
(d) The Wilcoxon signed-rank test was used for statistical analysis.
The significance of the P-value threshold was set at 5%. All plans were performed by the same experienced physicist and improved to meet the clinical objectives.

| RESULTS
Breast volumes on CT1 and CT2 are summed up in Table 1 the external breast shape. The subject who had the most different conformation between the two CT had a smaller breast volume on the CT2 but the breast's shape was modified ( Fig. 4 shows a breast volume less falling into the inferior and external directions). Breasts during RT develop edema 16,17 and tend to move into the anterior direction.

3.A | Results for the VB-VMAT and NoVB-VMAT plans on CT1
The first analysis compared the dose distributions on CT1 between plans optimized with and without the virtual bolus method. No degradation of coverage of target volumes were observed (     and 93.1%, respectively. The differential for median values was 0.9% for VB-VMAT and 0.7% for Tan-IMRT. There was no significant difference between the median differential of target volume coverage between the two techniques (P = 1).
T A B L E 1 Breast volume and its variation between the two scans. We chose to study one of the worst-case scenarios with respect to complexity of treatment volume, namely the left breast with whole regional nodal irradiation. In this setting, the standard method   Table 4, we chose to keep doses to contralateral OAR at a very low level. This is indeed a major concern in our plan optimization, particularly for young patients, as it has been shown that second cancer risk is dose dependent and inversely related to patient's age at his first treatment. 24,25 We found that the D mean to the skin was higher in VB-VMAT compared to NoVB-VMAT in both CT1 and CT2 evaluations. Our results are higher than previously published data, 12 but comparison should take into account the lateral and craniocaudal limits of the skin volume. Moreover, differences may be explained by the uncertainties of calculation in the first mm. Indeed, TPS with collapsed cone algorithms, do not provide accurate dosimetry in the first millimeters. [26][27][28] As stated in the AAPM TG 176 report, the depth of the sensitive basal layer ranges from 0.05 to 0.4 mm deep. 29,30 ICRU and ICRP selected 0.07 mm as the reference depth for the skin. 31 Thus, measurements made at an effective depth greater than the basal layer depth (such as 5 mm in this study or recent publication 12 ) will overestimate the "skin dose." The accuracy of superficial dosimetry depends on the dose calculation grid size. In this study, the grid size was 3 mm which is much bigger than the reference depth for the skin.

Patients
The clinical impact of such technique will still need to be evaluated. Studies reporting the clinical results of VMAT-based breast treatment demonstrated low toxicity with optimal local control. [6][7][8][9][10] The study reporting the outcome of stage III breast cancer treated with VMAT including the IMC in locoregional nodes reported higher doses to ipsi-and contralateral OAR than ours; however, their 2-year toxicity was low with no severe cardiac nor lung toxicities. 11 Prospective studies with toxicity analysis and long-term follow-up are needed.
The main limitation of this study is the relative small number of cases. However, to our knowledge our study is the first to report VMAT improves dose homogeneity and conformity for locoregional breast radiotherapy, but, due to the small and complex shape of the segments, their dose distribution may suffer more from intraand interfraction motion. In this study, using a virtual bolus, we focused on taking into account interfraction movements. Indeed, a systematic review covering 3378 studies concluded that interfraction motion is larger than intrafraction 32 which was confirmed in another recent study. 33 Interfraction motions are due to setup errors (random component) and conformation modifications (systematic component). Concerning setup errors, improving contention has always been a concern in radiation oncology: two studies 34,35 concluded that the use of a "Posi-Rest" minimized the setup errors by 10 mm, whereas another team only report the more comfortable aspect of a personalized contention. 36  F I G . 5. Dependency of ipsilateral lung and heart doses. This figure illustrates a potential dependency of heart dose and ipsilateral lung dose for regional left-sided breast VMAT RT. The four first studies (on the left) demonstrate V20 Gy < 25% for ipsilateral lung and an average dose (D mean ) to the heart >9 Gy. In the three other studies (on the right), the protection of the heart is improved (D mean < 6.5 Gy), but is accompanied by an increased the dose to ipsilateral lung.
bolus significantly improves the coverage of CTVs during the treatment fraction compared to technique which does not use it. Similar and even slightly better dose distribution are obtained when using a virtual bolus during the planning process. On a logistic level, although requiring two stages in planning, this technique is less timeconsuming than the previous standard field-in-field technique and is used in routine practice in our department for breast treatment including locoregional lymph nodes for patients over 50 years old.

CONF LICT OF I NTEREST
The authors declare no conflict of interest.