Dosimetric effects of anatomical deformations and positioning errors in VMAT breast radiotherapy

Abstract Aim Traditional radiotherapy treatment techniques of the breast are insensitive for deformations and swelling of the soft tissue. The purpose of this study was to evaluate the dose changes seen with tissue deformations using different image matching methods when VMAT technique was used, and compare these with tangential technique. Methods The study included 24 patients with breast or chest wall irradiations, nine of whom were bilateral. In addition to planar kV setup imaging, patients underwent weekly cone‐beam computed tomography (CBCT) imaging to evaluate soft tissue deformations. The effect of the deformations was evaluated on VMAT plans optimized with 5‐mm virtual bolus to create skin flash, and compared to standard tangential plans with 2.5 cm skin flash. Isocenter positioning using 2D imaging and CBCT were compared. Results With postural changes and soft tissue deformations, the target coverage decreased more in the VMAT plans than in the tangential plans. The planned V90% coverage was 98.3% and 99.0% in the tangential and VMAT plans, respectively. When tattoo‐based setup and online 2D match were used, the coverage decreased to 97.9% in tangential and 96.5% in VMAT plans (P < 0.001). With automatic CBCT‐based image match the respective coverages were 98.3% and 98.8%. In the cases of large soft tissue deformations, the replanning was needed for the VMAT plan, whereas the tangential plan still covered the whole target volume. Conclusions The skin flash created using an optimization bolus for VMAT plans was in most cases enough to take into account the soft tissue deformations seen in breast VMAT treatments. However, in some cases larger skin flash or replanning were needed. The use of 2D match decreased the target coverage for VMAT plans but not for FinF plans when compared to 3D match. The use of CBCT match is recommended when treating breast/chest wall patients with VMAT technique.


| INTRODUCTION
Adjuvant radiotherapy (RT) is recommended for breast cancer patients after breast-conserving surgery or node-positive (N+) mastectomy. 1 For a variety of treatment sites, such as prostate, brain, or head and neck, volumetric-modulated arc therapy (VMAT) treatment technique has been implemented rapidly into practice. 2 However, the RT of the breast is still largely accomplished with tangential field technique; not only due to the sparing of the contralateral breast from small doses caused by beam tails but also due to the possibility to account for possible swelling or deformation of the breast tissue during the treatment course using large field spillage outside the skin contour.
The VMAT has been shown to be a feasible treatment option for adjuvant RT of the breast. [3][4][5][6][7][8][9] It has been shown to reduce the dose to the ipsilateral lung for both left-and right-sided treatments, and to the heart for left-sided targets. 5,6 However, in VMAT the allowance for tissue deformations is limited. In the Monaco treatment planning system (Elekta AB, Stockholm, Sweden), the skin flash tool is implemented inside the VMAT plan optimizer. 5 In tomotherapy (TomoTherapy Hi-Art system, Madison, USA), the optimization of the dose to the skin is possible, but to allow for swelling, further extension of the field fluences requires additional bolus structures. 10 In the Eclipse treatment planning system (Varian Medical Systems, Palo Alto, CA, USA), there is currently no built-in tool to account for skin spillage in the VMAT planning and some clinics use a virtual bolus on the skin surface to create the skin flash. 4,6 The virtual bolus (0.5-1.0 cm thick) is only used in the VMAT optimization phase at locations where the planning target volume (PTV) is contoured to reach the skin surface with sufficient margin, and the bolus is removed for the final dose calculation. 4,6 The purpose of this study was to investigate the dosimetric effects of swelling, shrinking or other deformation of the breast or chest wall present in the RT treatment. The deformation was determined from weekly cone-beam computed tomography (CBCT) images collected from the breast VMAT treatments. The dosimetric effects of the deformations were investigated on the planning target volume (PTV) and heart dose parameters for different patient setups.

2.A | Patients
The inclusion criteria to this study were women being treated for breast cancer with adjuvant RT, and the breast or chest wall being treated with VMAT technique. In our clinic, patients are treated with VMAT if the dose constrains to the lung or heart are not met with sufficient PTV coverage using tangential half-blocked fields with the field in field technique (FinF). The planning criteria are less than 30% of the ipsilateral lung volume receiving 20 Gy; the mean dose to the ipsilateral lung being less than 15 Gy; and the mean dose to the heart being less than 3-5 Gy. Additionally, all bilateral breast or chest wall treatments are treated with VMAT due to better dose conformity.
Twenty-four patients were included in the study, as described in For all patients, the VMAT plans were designed as described by Boman et al. 6 Thus, the VMAT plans consisted of four partial arcs or, in case of bilateral treatment, of eight partial arcs. 6 Based on the beam's eye view images, the arcs were designed to avoid the lung, the heart, and the contralateral breast. The collimator angles (CA) in left-sided plans were between 10°and 30°for the anterior partial arcs, and complement angles were used for the lateral partial arcs. 6 For right-sided plans, the opposite angles were used. The field size was restricted to 15-18 cm in the left-right direction, that is, direction of MLC movement, to allow for optimal MLC modulation. As a modification to the split-arc design described in literature, 6 in some cases a small gap of 10°-20°between the frontal and lateral subarcs was used. This aided in minimizing the heart dose in right-sided cases. A virtual water-equivalent bolus of 5 mm was used in the areas where the PTV extended to the skin, but the final dose was calculated without the bolus, using normalization of mean dose of 100% to the PTV cropped 5 mm inside from the body contour (PTV-5 mm).
All patients selected to this study had traditional half-blocked FinF plans, which were used as a reference in dosimetric analyses.
The FinF plans consisted of two tangential half-blocked fields for the breast or chest wall, and three fields for the supraclavicular lymph-node region, including two anterior and one posterior field. 11 The tangential fields were extended laterally to the air by 2.5 cm to create the skin flash.

2.C | kV and CBCT imaging
Tattoo-based patient setup and 2D kV image guidance was used for all patients. [11][12][13] At least daily tangential and weekly orthogonal kV were imaged for the DIBH patients, and weekly tangential kV for the free breathing patients. [11][12][13] The tolerances for chest wall were (1) The CBCT-based body contour was used in creating outside and inside structures with Boolean subtraction operators, where the BODY contour in the CBCT was extending outside or inside of the BODY contour of the original CT, respectively.
(2) The outside structure consisted of the swollen tissue and was assigned to −100 HU as in fat tissue.
(3) The inside structure consisted of areas of tissue shrinkage and was assigned to −1000 HU as in the air. The tolerances for chest wall and heart were 5 mm in C-C and 4 mm in lateral and A-P directions. 12,13 The three dose distributions based on different matching techniques (3D + rot, 2D + rot, 2D) were calculated for both the VMAT and FinF plans, resulting in six dose distributions for each CBCT image.

2.D | Image analysis
For each surface-corrected structure set, the heart and humeral head(s) were modified from the planning CT to match the CBCT image.
Additionally, the PTV border was modified to match the skin surface seen in the CBCT image. An experienced oncologist reviewed all modifications. For nine patients the humeral head was not shown in the CBCT due to the limited image field of view (FOV) of 16 cm in the C-C direction, and only 16 patients were included in the analyses of humeral head. Also the PTV and heart were partially cut out due to the restricted FOV in CBCT, but their position was evaluated based on the part shown on CBCT. In order to minimize the uncertainties caused by the cranially missing CBCT data, PTV-5 mm was evaluated also by dividing it into the breast or chest wall (PTVb/c) and the supraclavicular (PTVsclav) regions. The junction between PTVb/c and PTVsclav regions was made on the CT slice where the original PTV was no longer extended to the skin (Fig. 1). The PTVb/c was therefore mostly unaffected by missing CBCT data and was used, where V95(PTV) (cc) was the volume which receives at least 95% of the prescribed dose in the PTV, V(PTV)(cc)was the total volume of PTV, and V95(cc) was the volume of the whole body which received at least 95% of the prescribed dose. For the HI, ribs. 12 For the DIBH patients, the residual error in breath hold level (BHL) was measured on the CBCT images in both A-P and C-C directions using the difference between two independent CT-CBCT registrations: first matching the vertebra, and then matching the sternum. 14

3.A | Original plan quality
In the original plans, the HI and CI were better in the VMAT plans than in the FinF plans (Fig. 3). For the CI, the differences were more pronounced. The differences in other DVH parameters of the PTV were higher V90%(PTV) and V95%(PTV) except for the clavicular region; and lower V105% and D2 cc in the VMAT than in the FinF plans (Fig. 3). The D2% of the heart was lower in the VMAT than in the FinF plans. Otherwise the heart doses were closely similar, with large doses going to a slightly larger volume in the FinF plans, and with low doses spreading to a slightly larger volume in the VMAT plans. For the humeral head, the VMAT doses were significantly lower than the FinF doses.

3.B | Tissue deformation
The range of tissue deformations d s in the tangential images (Fig. 2) was from −5 to 27 mm, with median swelling of 2 mm and interquartile range from 0 to 4 mm. No correlation was found between increasing swelling and time from the first fraction (Spearman's rho = −0.109). Instead, some patients had swelling in the beginning, some at the end of the treatment course, and several had no or only minor swelling, shrinkage, or deformation. The correlations between skin deformations d s (Fig. 2)

3.C | Isocenter error
For all FinF and VMAT recalculated plans (3D + rot, 2D + rot and 2D) the HI and CI indicated reduced plan quality when compared to the original plans. Differences between the original and recalculated plans indicated consistently that the 3D + rot match was closest to the original plan, for 2D + rot and 2D match the differences were slightly larger for both the VMAT and the FinF plans [ Fig. 3(a)]. The changes in HI and CI from the original plan to the recalculated plans The maximum dose D2 cc increased when patient setup inaccuracies were introduced, but it was not dependent on the matching method.
The individual P-values for the differences between plan and different matching methods are presented in Table 3 for the dose minima and Table 4 for the dose maxima. for all patients and fractions (Fig. 4), except for one fraction of Pat #17 with the 2.7-mm seroma. In the FinF plans the V90% difference between 2D and 3D + rot match was small (Fig. 4). If the 2D match was used, the FinF plans provided better PTV coverage (P < 0.001).
When the 3D + rot match was used, VMAT proved slightly better in PTV coverage (P < 0.001). The difference between 2D and 2D + rot was negligible.
The correlations between the dose parameters of the PTV coverage and isocenter errors in the online 2D match in the A-P, C-C, and lateral direction, and in the patient rotation were modest at maximum for both FinF and VMAT recalculated plans. The HI(PTV-5 mm) was modestly correlated with the A-P isocenter error (Spearman's rho = 0.510) in the VMAT plans. This was accompanied by weak correlations (ρ = 0.330-0.433) in V95%, V105%, and D2 cc. In the FinF plans, the individual parameters were weakly correlated with the A-P isocenter error (ρ < 0.5), but neither HI nor CI were affected by these.

3.D | Heart and humeral head
The doses to the heart and humeral head are shown in Fig. 3(e)- Table 5. The V20%, V10%, V5% and the mean dose to the heart increased slightly in the FinF recalculated plans (3D + rot, 2D + rot and 2D) but decreased in the VMAT The lateral and A-P residual errors of the vertebrae, sternum, and ribs in the 2D match were not significantly correlated with the heart dose change, and nor were the residual errors of ribs in the tangential 2D images.

3(f), and the P-values in
In the DIBH patients, small to moderate changes in the BHL were seen in the A-P direction on the CBCT images (median 0 cm, range −0.8…0.8 cm, interquartile range −0.3…+0.2 cm).
These changes correlated modestly with all heart dose parameters T A B L E 2 Correlation coefficients for correlations between the tissue deformations d s (Fig. 2) measured in tangential images and DVH changes in PTV dose minima (V90% and V95%) and maxima (V105% and D2 cc). Values are bolded where P < 0.05. Spearman's rho test. T A B L E 3 P-values for differences in the dose minima V90%(%) and V95%(%). Pairwise comparisons are performed first as differences from the original plan to actual CBCT-based patient geometry using each of the three matching techniques, and second between the three matching techniques. Statistically significant values are in bold.

4.C | Heart dose
A small isocenter error in the A-P direction correlated with the heart dose change. This was expected as the high-dose volume got closer to or further from the heart. In addition, correlation was found between the heart dose and the C-C residual error of the spine. This was likely due to the effect of C-C breathing motion. If the A-P BHL is lower than planned, it is more likely that positional errors also occur in the C-C direction. 14 Thus, if in 2D image guidance the match of lateral image is performed with a compromise between the sternum and spine, but prioritizing the sternum, then the C-C residual error of the spine is likely connected with the altered breathing motion. Indeed, we have seen several patients whose breathing movement is equally or even more pronounced in the C-C than in the A-P direction. This finding was supported by the residual error in the BHL of DIBH patients, where both the A-P and C-C error affected the heart dose. Therefore, attention should be paid not only on the A-P but also on the C-C breathing motion. Errors in the BHL should be corrected by adjusting the BHL limits and/or instructing the patient based on visual assessment of the lateral image.

4.D | Supraclavicular region
The accuracy of the supraclavicular region (PTVsclav) was limited by the CBCT image size and should only be considered for isocenter positioning errors, not anatomical deformations. The PTV contours could not be verified outside the CBCT image range. Due to the optimization criteria, the irradiation of the supraclavicular regions traverses primarily through the A-P directions in VMAT treatments.
However, unlike in traditional planning, part of the irradiation enters the patient from the lateral direction. The inclusion of lateral beams increases the uncertainty of planned dose delivery if the shoulder moves especially in the C-C or A-P direction. 16

4.E | Imaging protocols
Using Varian OBI low-dose thorax CBCT mode, the CBCT-induced dose to the breast is in the range of 0.4-1 cGy per fraction. [17][18][19] The dose to the heart is in the range of 0.4-1 cGy, and to the lung 0.5-0.7 cGy. 17,18,20 Even though these are not large doses, the cumulative dose of daily CBCT does increase the doses to the OARs compared to 2D kV imaging. However, this might be compensated by the decreased heart doses with the more accurate setup. Furthermore, the doses are nearly halved in Varian TrueBeam CBCT. 20 The modification of the imaging protocols will decrease the doses also in Instead, for the VMAT plans the 3D + rot match was seen superior. The CBCT match is therefore recommended to be used as frequently as possible (e.g., daily) for patient positioning in VMAT treatments both to decrease the average error, and to eliminate large random errors.

| CONCLUSION
Considering the soft tissue deformations and breast tissue swelling during the course of radiotherapy of the breast, the actual dosimetric parameters would be similar to the plan for both VMAT and FinF plans if 3D image matching was used. Especially for VMAT plans, the changes in the dosimetric parameters became worse with the online 2D matching methods. In VMAT treatments, a daily automatic CBCT matching along with monitoring of the skin surface is recommended. Large deformations in any direction should be evaluated in terms of their clinical relevance, and the potential need for replanning should be investigated.