Dosimetric variation in preoperative partial breast radiosurgery assessed by deformable image registrations

Abstract

Objective

To assess dosimetric variation caused by breast deformation in breast radiosurgery based on deformable image registration.

Methods

This study included 30 patients who were treated in the prone position for preoperative partial breast radiosurgery. The biopsy clip in CBCT was aligned to the one from the planning CT. Deformable image registration (DIR) was performed to deform the planning CT into the CBCT, focusing on the breast shape. The treated plan (P_Tx) was recalculated based on the deformed CT. Thus, P_Tx represented the actual treatment delivered to the patient and was compared to the original plan (P_Org).

Results

The mean differences of target volumes covered by 95% and 100% of the prescribed dose between P_Org and P_Tx were less than 0.5%. The mean differences ± standard division for skin maximum dose (D_max), dose to 1cc (D_1cc) and D_10cc were 0.3 ± 0.7 Gy, 0.3 ± 0.6 Gy and 0.6 ± 0.6Gy between P_Org and P_Tx, respectively.

Conclusion

The treated plan was accurately recalculated based on the deformed CT. Despite slight variance in breast deformation, the dosimetric variation was very small, ensuring that adequate target coverage and skin dose were maintained during treatment as planned originally.

INTRODUCTION

Postoperative partial breast irradiation (PBI) using an external beam technique has been commonly used to treat low-risk early breast cancer patients with a smaller irradiated volume and larger fractional dose than standard whole breast irradiation (WBI).^1,2 To further improve treatment efficiency, feasibility of single-fraction partial breast radiosurgery (PBRS) has been investigated and shown promising early results.^3-6 Treatment of the intact tumor preoperatively allows a significant reduction of treatment volume with a more accurately defined high-risk target area than a postoperative seroma.^7,8 Based on such studies, the institutional review board (IRB) at our institution approved phase I and phase II clinical trials to evaluate preoperative PBRS for early stage breast cancer patients.

Both PBRS phase I and phase II studies had the primary objective to report cosmetic outcome with the hypothesis that cosmesis would be improved by sparing skin and normal breast tissue with the reduced target volume. Patients were treated in the prone position to minimize lung, heart and contralateral breast dose, as well as to separate the target from the chest wall.⁴ Other advantages of prone positioning included more favorable tumor to skin position and image fusion of standard prone position breast MRI for target delineation.^5,9 However, prone breast position associated with more setup variability and breast deformation though with minimal respiratory motion.^10-12 This can potentially lead to undesirable dosimetric variation in treatment delivery from the original treatment plan. With intensive image-guided radiation therapy (IGRT) technique and biopsy clip as a surrogate of the target, accurate target localization can be achieved.^{11, 13}

Nevertheless, breast shapes were noted to be different between planning CT and treatment CBCT and the degree of dosimetric variation caused by such breast deformation was unknown. This study assessed dosimetric variation caused by breast deformation in PBRS treatment based on dose recalculation using deformable image registration (DIR). The deformation was determined between planning CT and treatment CBCT. The dosimetric variation was evaluated for target coverage and skin dose. This evaluation could be used for assessing the need for real-time adaptive treatment and provide insight into dosimetric factors that might impact cosmetic outcome.

MATERIALS AND METHODS

Treatment planning

This study included 30 patients who were enrolled in the IRB approved trial phase II study from 2017 to 2021. Patients underwent planning CT scan with 1 mm slice thickness in the prone position on a CDR prone breast board (CDR systems Inc, Calgary, Alberta, Canada). A contrast-enhanced treatment planning MRI was acquired in the prone position on a dedicated breast surface coil. CT and MR images were imported into the Eclipse treatment planning system (TPS) (Varian Medial Systems, Inc., Palo Alto, CA) and registered to align the biopsy clip as all patients had one clip implanted in the tumor during the biopsy and soft tissue around the tumor using manual rigid-body registration. Gross tumor volume (GTV) was delineated by the attending physician with input from a radiologist specializing in breast imaging when needed. Clinical target volume (CTV) was created with a 1.5 cm margin from the GTV. Planning target volume (PTV) based on CTV (PTV_CTV) had a 0.5 cm margin from the CTV and 15 Gy (Rx_CTV) in a single fraction was prescribed. PTV based on GTV (PTV_GTV) had a 0.5 cm margin from the GTV and 21 Gy (Rx_GTV) in a single fraction was prescribed as a simultaneous integrated boost. The PTV_CTV was modified to crop 5 mm in from the skin surface.^{1, 11} Physicians manually modified the PTV_CTV near the chest wall and/or skin if needed. Skin was defined as a 3 mm layer along the skin surface.⁴ Heart, lungs, ipsilateral breast and contralateral breast were contoured.

Intensity modulated radiation therapy (IMRT) was used to achieve optimal treatment plans using 4 to 7 co-planar beams with minimum of 10° to 15° separation in between beams with 6 MV flattening filter free (FFF) photons on a TrueBeam linear accelerator machine (Varian Medial Systems, Inc., Palo Alto, CA).^{14, 15} Planners manually set gantry angle and field size for each beam to minimize dose to the heart and to avoid the contralateral breast based on beam’s eye view (BEV). Due to the patient size and the locations of the heart and contralateral breast in relation to the target, some cases could only fit 4 beams. Optimization was performed to achieve target coverage objectives and normal tissue constraints listed in Table 1. Dose calculation was performed using Eclipse Analytic Anisotropic Algorithm (AAA) v13.5.23 or v15.6.03 with 2.5 mm dose calculation grid.

Table 1.

Dose and volume objectives and constraints used for optimization

Structure	Dose or Volume	Objectives/Constraints
PTVGTV	V19.95Gy (=95% of RxGTV)	≥95%
PTVCTV	V14.25Gy (=95% of RxCTV)	≥90%
CTV	V15Gy (=100% of RxCTV)	≥95%
Skin	D max	≤21 Gy
	D 1cc	≤14 Gy
	D 10cc	≤9 Gy
Heart	D mean	≤1.5 Gy
Lungs	D mean	≤3.6 Gy
Ipsilateral breast	D 30%	≤10.5 Gy
contralateral breast	D max	≤2.1 Gy

V_xGy is the percentage volume receiving at least x Gy, D_max is the maximum dose, D_mean is the mean dose, D_xcc is the absolute dose to x cc, and D_x% is the absolute dose to x %.

Patient Setup

In the treatment room, the patient was first aligned based on the skin marks made during the planning CT scan. Orthogonal 2D kV images were taken to confirm the general agreement of bony anatomy such as chest walls, spine, clavicle, and arm between kV images and digitally reconstructed radiographs (DRR). The couch was shifted to align the biopsy clip in acquired kV images and DRR. Cone-beam CT (CBCT) images were taken to verify the biopsy clip and surrounding soft tissue alignment in 3D display. If the breast shape or overall patient positioning on the CBCT did not match with the planning CT based on visual inspection and if the discrepancy was affecting the beam entries, therapists attempted to re-position the patient in the room and repeated the CBCT. KV images were taken again just before the first treatment beam and in-between treatment beams to minimize intrafactional motion.¹³ The kV image was taken when the kV projection was lateral or lateral oblique to clearly acquire the clip. If the biopsy clip position was off more than 3 mm, the couch was shifted.

Deformable image registration

DIR was performed to deform the planning CT into the treatment CBCT to create a deformed CT (CT_Def). Thus, CT_Def had the image quality equivalent to the planning CT while representing the actual patient position and breast shape from CBCT acquired at the time of treatment. MIM software (version 6.9.4; MIM Software Inc, Cleveland, OH) was used for DIR using the CBCT as the reference image and the planning CT as the moving image. A rigid registration was first performed in MIM with a box region of interest that encompassed the ipsilateral breast and chest wall, followed by a free-form DIR using the normalized intensity-based algorithm provided in the software. The initial DIR result was further improved using the Reg Refine tool by locking alignments of a few landmark points including the biopsy clip and points at tissue-air interface ¹⁶. The ipsilateral breast contour on the planning CT was accordingly transformed to a CT_Def by applying same deformation from final DIR result. The CT_Def and the ipsilateral breast contour were then transferred into the Eclipse TPS for validation and dose calculation.

Deformable image registration validation

The ipsilateral breast contour on the CT was cropped to include the breast only along the longitudinal ranged 1 cm beyond the PTV_CTV and labeled as a breast volume on the planning CT (V_CT). V_CT was copied to the treatment CBCT based on the rigid alignment performed at the time of treatment and relabeled as V_CBCT to define the breast volume on the CBCT (V_CBCT). V_CBCT was manual modified along the body surface and chestwall area as the breast shape was different between the CT and CBCT. Therefore, V_CBCT represented the breast on the CBCT contoured in the same way as the V_CT represented the breast on the CT. The deformed ipsilateral breast contour on the CT_Def was cropped to contain the same longitudinal range as V_CT and it was labeled as V_CTDef to define the deformed breast volume. A contour-based metric, Dice similarity coefficient (DSC), can quantify overlaps of two structures.¹⁷ The degree of breast deformation between the planning CT and the treatment CBCT was quantified by DSC_Def = 2 (V_CT ∩ V_CBCT)/(V_CT + V_CBCT). A DSC_Def close to 1 would indicate a high degree of overlap between the CT and CBCT, therefore, inferring negligible deformation from simulation to treatment. The accuracy of DIR was quantitatively validated by DSC_DIR= 2 (V_CTDef ∩ V_CBCT)/(V_CTDef + V_CBCT). A DSC_DIR close to 1 would indicate a high degree of overlap between the CBCT and CT_Def, therefore, validating accurate DIR performance in this study.

Dosimetric evaluation

The CT_Def preserved Hounsfield Unit (HU) values from the planning CT while it represented actual patient position and breast shape from CBCT acquired at the time of treatment. The skin structure was re-generated on the CT_Def because the deformed skin did not have a consistent 3 mm layer along the skin surface. The body outline and ipsilateral breast structure were also re-generated following the deformed body outline. The GTV was copied from the planning CT to the CT_Def with an assumption that the GTV was not deformed given very small size and its fixed location at the biopsy clip. The CTV, PTV_GTV and PTV_CTV were newly generated on the CT_Def as those were not anatomical-based structures but geometrical margin-based structures expanded from the GTV. The original plan (P_Org) based on the planning CT was recalculated based on the CT_Def to create the treated plan (P_Tx) with the fixed Mus and MLC sequence. Dosimetric parameters consisting of PTV_GTV V_19.95Gy (=95% of Rx_GTV), PTV_CTV V_14.25Gy (=95% of Rx_CTV), CTV V_15Gy (=100% of Rx_CTV), and D_max, D_1cc and D_10cc for skin were compared between two plans. The statistical significance was determined with a 2-tailed paired t test (P value) with a significance threshold of 0.01 (P < 0.01).

RESULTS

Table 2 shows the volumes of structures and the comparison of dosimetric parameters between P_Org and P_Tx. The original plans (P_Org) satisfied all of the dose objectives and constraints shown in Table 1. Skin volume is not included in the table because its volume varies depending on the range of region of interest selected to contour the skin and the absolute volume of skin near the target and the beam paths is the parameter of interest for this study. The mean differences between P_Org and P_Tx for target coverage parameters were very small and statistically insignificant. The mean differences between P_Org and P_Tx for skin doses were also small, but statistically significant for skin D_1cc and D_10cc. As explained earlier, the percentage volume could not be quantified for the deformed ipsilateral breast because of the limited CBCT FOV. Thus, Table 2 only included the ipsilateral breast D_30% for P_org.

Table 2.

Comparison of mean dosimetric parameters ± standard deviation between P_Org and P_Tx

Structure	Dose or volume	Objectives/ constraints	Volume in CT [cc]	Volume in CTDef [cc]	POrg	PTx	Difference
PTVGTV	V19.95Gy (=95% of RxGTV) [%]	≥95%	6.2 ± 2.3	5.9 ± 2.2	99.5 ± 1.0%	99.6 ± 0.8%	0.1 ± 0.6%
PTVCTV	V14.25Gy (=95% of RxCTV) [%]	≥90%	72.4 ± 15.9	70.4 ± 17.7	98.4 ± 2.1%	98.0 ± 2.2%	−0.4 ± 1.2%
>CTV	V>15Gy (=100% of Rx>CTV) [%]	≥95%	39.4 ± 8.8	39.2 ± 8.6	99.3 ± 0.8%	99.0 ± 1.4%	−0.3 ± 1.0%
Skin	D>max [Gy]	≤21 Gy	NA	NA	14.7 ± 1.7 Gy	15.0 ± 1.6 Gy	0.3 ± 0.7 Gy
	D1cc [Gy]	≤14 Gy			12.1 ± 1.3 Gy	12.4 ± 1.3 Gy	0.3 ± 0.6 Gy*
	D10cc [Gy]	≤9 Gy			7.4 ± 1.2 Gy	8.0 ± 1.2 Gy	0.6 ± 0.6 Gy*
Ipsilateral breast	D>30% [Gy]	≤10.5 Gy	1257.3 ± 490.8	NA	3.4 ± 1.8 Gy	NA	NA

Difference = P_Tx – P_Ori. Positive difference indicates a higher dose delivered for the treatment. Negative difference indicates a lower dose delivered for the treatment.

*p < 0.01

Cropped breast is V_>CT in CT and P_>Org, and V_>CTDef in CT_Def and P_>Tx.

NA: Not applicable.

Figure 1-(a) shows boxplots of percentage volume differences between P_Org and P_Tx for PTV_GTV V_19.95Gy, PTV_CTV V_14.25Gy, and CTV V_15Gy and Figure 1-(b) shows boxplots of absolute dose differences for skin D_max, D_1cc and D_10cc. In general, target coverage was reduced and skin dose was increased in the treated plans compared to the original plans by small differences.

Figure 1.

Boxplots of percentage volume differences between P_Org and P_Tx for PTV_GTV V_19.95Gy, PTV_CTV V_14.25Gy, and CTV V_15Gy in a) and absolute dose differences between P_Org and P_Tx for Skin D_max, D_1cc and D_10cc in b).

The degree of deformation and DIR accuracy are displayed in Figure 2. The x-axis is the DSC_Def that represents the degree of deformation obtained from the overlap of breast contours between the planning CT and treatment CBCT for each patient. The greater DSC_Def implies less of deformation. The y-axis is the DSC_DIR that represents the accuracy of the DIR obtained from the overlap of breast contours between the CBCT and CT_Def. The greater DSC_DIR implies more accurate DIR. The mean ± standard deviation for DSC_Def was 0.91 ± 0.03 with the range of 0.86 to 0.96 and was 0.98 ± 0.01 for DSC_DIR with the range of 0.96 and 1.00. No correlation between DSC_Def and DSC_DIR was found. We also confirmed that DSC_Def had no correlation with the breast volume.

Figure 2.

Scatter plot of DSC_Def and DSC_DIR where DSC_Def (x-axis) indicates the degree of deformation and DSC_DIR (y-axis) indicates the deformable image registration accuracy.

Figure 3 shows an example patient with a relatively large deformation (DSC_Def=0.86) and relatively less accurate DIR (DSC_DIR=0.97) compared to other cases. This patient is indicated with the red circle in Figure 2. Figure 3 (a) and (b) are the blended images of planning CT and CBCT with V_CT and V_CBCT contours in blue and purple, respectively. The breast shapes on the planning CT and the treatment CBCT appear different, which leads to a relatively small DSC value representing a relatively large breast deformation between simulation and treatment. Nevertheless, such discrepancy did not directly affect the skin dose variation. We found no correlation between DSC_Def and skin dose parameters. Figure 3 (c) and (d) are the blended images of CBCT and CT_Def with V_CBCT and V_CTDef contours in purple and cyan, respectively. The two structures overlap very well except for the posterior-lateral portion of the breast, which led to the relatively smaller DSC value representing a less accurate DIR compared to other patients in this study. Most patients who had a DSC_DIR near 1.0 had good overlap of V_CBCT and V_CTDef throughout the breast and those with a DSC_DIR value close to 0.96 showed small discrepancy mostly around the posterior-lateral area of the breast between V_CBCT and V_CTDef.

Figure 3.

Blended images of CT and CBCT in a) axial view and b) sagittal view with contours of V_CT and V_CBCT in blue and purple, respectively. Blended images of CBCT and CTDef in c) axial view and d) sagittal view with contours of V_CBCT and V_CTDef in purple and cyan, respectively. GTV, CTV and PTV_CTV are contoured in red, pink and orange, respectively.

DISCUSSION

Accurate translational patient positioning was achieved for PBRS by aligning the biopsy clip as a surrogate for the tumor for treatment.¹³ Nevertheless, the effect of the breast deformation on delivered treatment dose compared to the planned dose was unknown. This study was designed to evaluate dosimetric variation in delivered treatment dose due to breast deformation. For whole breast treatment, similar studies have been performed. Zegers et al. studied the dosimetric effect of breast deformation for supine position whole breast 3D plans.¹⁸ Their study used re-scanned CT images instead of treatment CBCT because of the limitation in field-of-view (FOV) and inaccurate HU of CBCT. Rossi et al. aimed to quantify the dosimetric effect of breast deformation for breast VMAT.¹⁹ In order to maintain dose calculation accuracy, they used the original CT with the body contour modified for CBCT. Proper HU values were assigned for swollen or shrunken tissue areas. Xiao et al. simply simulated dosimetric effect from the translation setup errors in prone position by recalculating treatment plan based on planning CT because of CBCT artifacts and FOV limitation.²⁰ Their study assumed minimal effect of breast deformation for 3D CRT. Due to inaccurate HU values and limited FOV, direct utilization of CBCT for dose calculation could yield inaccurate results.^{21, 22} We designed this study to utilize the original CT, but deformed to the treatment CBCT. Therefore, the CT_Def maintained accurate HU values from the planning CT with the body outline from the treatment CBCT that represented the breast shape at the time of treatment.

Similar studies for various treatment sites were also performed. Ideally, dose calculation based on CBCT can represent the actual delivered dose to the patient. Studies have used CBCT-based dose calculation with HU correction or improvement of HU accuracy of CBCT images through different hardware/software approaches.^{23, 24} Generating a pseudo CBCT with more accurate HU numbers based on deep learning has been investigated to improve dose calculation accuracy.²⁵ With high accuracy of DIR, which was the case in this study, the DIR based method is preferred as the CT_Def preserves HU values from the planning CT while maintaining the same anatomical shape as the CBCT that represents the treatment.²⁶

We used MIM software to perform DIR in this study. MIM used free-form deformation algorithm and different similarity metrics for optimization depending on the modality of images to be registered together. The normalized intensity-based registration was used as it’s recommended by the software for DIR between CT and CBCT. Several studies have evaluated the accuracy of MIM DIR performance using virtual or anthropomorphic phantoms for various disease sites.^{17, 27-29} MIM consistently showed equivalent or better performance compared to other software systems in multi-vendor comparison studies.^{17, 27, 29} This study showed the feasibility of MIM performance for DIR of CT to CBCT for prone breast cases. We achieved good DIR in matching different breast shapes as supported by high DSC scores as well as verification with visual inspection. Small discrepancy in DIR was found mainly in posterior and/or posterior-lateral portion of the breast in visual inspection. The anterior and anterior-lateral outline of the breast that affected dosimetric variation as beams entered through was well registered. It implies that the dose recalculated based on CT_Def using MIM accurately represent the actual dose delivered to the patient.

Doses to heart, lungs, ipsilateral breast and contralateral breast were not evaluated in this study. The DIR focused solely on breast deformation that could potentially affect treatment delivery. Deformation of heart, lung, or ipsilateral breast was not considered because those structures did not affect beam entry, consequently, not causing dosimetric variation to the target or skin. Additionally, given the limited FOV of CBCT, those structures were only partially imaged. Therefore, the mean dose or percentage volume of those structures were not quantifiable. Hence, this study focused on evaluation of doses to the targets and skin.

Our PBRS utilized patient setup, treatment planning and delivery techniques that are very similar to typical IMRT PBI in prone position.^30-32 A dose calculation grid size of 2.5 mm was used due to large calculation volume and similarity with whole or partial breast irradiation planning. However, a finer dose calculation grid (≤2 mm) shall be considered for future practice as it’s recommended by AAPM Task Group Report 101.³³ There was no special immobilization utilized for our PBRS as the patient setup was similar to typical whole or partial breast radiotherapy in prone position. However, intensive imaging was implemented to localize the target accurately and to correct intrafractional motion. The dose fall-off was controlled by limiting D_30% of ipsilateral breast similar to PBI^30,32 instead of typical dose gradient index used in brain radiosurgery.³⁴ All these similarities between our PBRS and typical PBI imply that many institutions have potential to launch breast radiosurgery program. Nevertheless, physicists and physicians should determine specific requirements in designing the treatment process to achieve accurate and safe treatment. The specific requirements we implemented are a biopsy clip as a target surrogate, MR images for target delineation, 1 mm CT slice thickness for clear display of the biopsy clip, skin dose constraints, and intensive imaging for patient setup throughout the treatment. Institutions may identify different or unique requirements that are essential for their institutions as they implement breast radiosurgery program.

This study evaluated the dosimetric variation caused by breast deformation in order to understand the delivered dose to the patients. The skin doses were slightly increased in the treated plans compared to the original plans in this study. The degree of deformation quantified as DSC_Def showed negligible correlation with changes in skin doses. The changes in target coverage also showed negligible correlation with the degree of deformation. We can deduce that the degree of deformation was not as much as it would affect accuracy of dose delivery and the multiple beam IMRT plan was robust for prone breast treatment even with some breast deformation. Rather, skin dose showed strong correlation between the original plans and the treated plans. Two cases for skin D_1cc and five cases for skin D_10cc had dose values greater than the constraints in P_Tx by less than 1 Gy. Those cases had skin D_1cc and D_10cc values close to the constraints in the original plans. Thus, a small variation triggered those parameters to be slightly greater than the constraints. In sum, our data show that patients well received the treatment as originally planned for PBRS in prone position, and that breast deformation did not significantly affect dose variation. Having a high quality initial plan and localizing the target based on the biopsy clip were critical components to achieve these successful treatments. Importantly, quantifying the actual delivered dose will allow better understanding of the relationship between cosmetic outcomes and radiation dose during longer term clinical follow up.

CONCLUSION

This study investigated the dosimetric variation caused by breast deformation between simulation and treatment for partial breast radiosurgery treatment in the prone position. The treated plan was recalculated based on the deformed CT with accurate HU values from the planning CT and body outline from the treatment CBCT that represented the breast shape at the time of treatment. Though deformation of the breast was observed, the dosimetric variation was very small, demonstrating that intended target coverage and skin sparing were maintained.