Modeling AeroForm tissue expander for postmastectomy radiation therapy

Abstract The AeroForm chest wall tissue expander (TE) is a silicon shell containing a metallic CO2 reservoir, placed surgically after mastectomy. The patient uses a remote control to release compressed CO2 from the reservoir to inflate the expander. AeroForm poses challenges in a radiation therapy setting: The high density of the metallic reservoir causes imaging artifacts on the planning CT, which encumber structure definition and cause misrepresentation of density information, in turn affecting dose calculation. Additionally, convolution‐based dose calculation algorithms may not be well‐suited to calculate dose in and around high‐density materials. In this study, a model of the AeroForm TE was created in Eclipse treatment planning system (TPS). The TPS model was validated by comparing measured to calculated transmission through the AeroForm. Transmission was measured with various geometries using radiochromic film. Dose was calculated with both Varian’s Anisotropic Analytical Algorithm (AAA) and Acuros External Beam (AXB) algorithms. AAA and AXB were compared using dose profile and gamma analyses. While both algorithms modeled direct transmission well, AXB better modeled lateral scatter from the AeroForm TE. Clinical significance was evaluated using clinical data from four patients with AeroForm TEs. The AeroForm TPS model was applied, and RT plans were optimized using AAA, then re‐calculated with AXB. Structures of clinical significance were defined and dose volume histogram analysis was performed. Compared to AXB, AAA overestimates dose in the AeroForm device. Changes in clinically significant regions were patient‐ and plan‐specific. This study proposes a clinical procedure for modeling the AeroForm in a commercial TPS, and discusses the limitations of dose calculation in and around the device. An understanding of dose calculation accuracy in the vicinity of the AeroForm is critical for assessing individual plan quality, appropriateness of different planning techniques and dose calculation algorithms, and even the decision to use the AeroForm in a postmastectomy radiation therapy setting.


| INTRODUCTION
For breast cancer patients with greater than four positive lymph nodes, positive or close margins, or a tumor greater than 5 cm, the National Comprehensive Cancer Network guidelines recommend postmastectomy radiation therapy (PMRT). Currently there is no consensus on the optimal method and timing of postmastectomy breast reconstruction for patients receiving PMRT. [1][2][3] One common technique is the two-stage breast reconstruction. In this method, a tissue expander (TE) is placed at the time of mastectomy. In the following weeks, the TE is gradually expanded to the desired size, and PMRT is delivered after expansion is complete. At a later time, the TE is exchanged for a permanent implant. [4][5][6] Several types of TEs are available. The CPX® (Mentor, Irvine, CA, USA), Natrelle® (ALLERGAN, Santa Barbara, CA, USA), Dermaspan TM and AlloX 2 ® (Sientra, INC., Santa Barbara, CA, USA) TEs consists of a silicon shell containing a magnetic injection port.
The TE is expanded by externally aligning a magnetic port locating device to the internal magnetic port (IMP). Once aligned, saline is injected percutaneously through the IMP to inflate the TE.
Patients receive weekly injections, over the course of 6-8 weeks. 7,8 The AeroForm TE (AirXpanders, Palo Alto, CA, USA) consists of a silicon shell containing a stainless steel reservoir of compressed CO 2 .
The TE is expanded using a hand-held remote control, which sends a radio-frequency signal to the expander. The TE in turn releases small volumes of compressed CO 2 from the reservoir into the silicon shell.
The expansion is patient controlled; typically a patient releases 10 cc CO 2 at a time, up to three times per day, over the course of 4-6 weeks. If radiation is indicated, the patient is typically simulated and treated after expansion is complete. 8,9 Tissue expanders pose particular challenges in a radiation oncology setting. The high-density metallic components of both types of TE cause artifacts in treatment planning CT images. The degradation of image quality makes target definition more challenging, and misrepresentation of CT number in and around the TE may cause errors in dose calculation. 16-bit CT reconstruction corrected using metal artifact reduction (MAR) reconstruction techniques may minimize the effects of artifacts on image quality. 10,11 However, even with accurate CT information, convolution dose calculation algorithms are not designed for use with high-Z materials. 12,13 Previous literature has proposed various solutions dealing with TEs in a radiation treatment planning systems (TPS). 14 Chen 17 Moni placed OSLDs at various locations around the Aero-Form TE on anthropomorphic phantom. The set-up was imaged and an RT plan was created and delivered. The accuracy of AAA with and without heterogeneity corrections was evaluated by comparing the calculated dose to the OSLD-measured dose. Moni reported agreement between AAA and measured data within 5%, except at the reservoir-chest wall-interface, where the measured dose was consistently 5-15% higher than calculated. Moni was also unable to observe the predicted "dose shadow" effect with OSLD measurements, possibly due to uncertainty in detector placement. 17 These areas are of significant clinical importance; clinicians require accurate dose calculation in these regions to assess overall RT plan quality.
In a 2014 study, Tran et al measured transmission through various components of the AeroForm CO 2 reservoir using an ion chamber. 18 The authors reported physical density assignments in Pinnacle TPS that resulted in dose calculations in agreement with measured transmission data. Significant differences existed between the experimentally determined density and vendor-reported density. Using an anthropomorphic phantom with the AeroForm TE, Tran et al compared dose calculated in Pinnacle TPS to dose measured with TLDs at various locations. Percent differences between calculated and measured doses ranged between −10 and +15%. 18 This study deals with the accuracy of modeling of the Aeroform TE in Eclipse TPS using both AAA and Acuros External Beam (AXB) algorithms (Varian Medical Systems, Palo Alto, CA, USA). AAA is under the general class of superposition/convolution algorithms in which total dose is computed by superposition/convolution of the primary dose with scatter dose kernels. Such algorithms account indirectly and approximately for electron transport in heterogeneous media. AAA uses a 3D pencil-beam kernel, which accounts for changes in electron density perpendicular to the beam direction by applying radiological depth scaling. 19 The AXB algorithm is considered a grid-based Boltzmann Solver (GBBS) as it analytically solves the Boltzmann transport equation using interaction cross sections specific to the relevant material and energy. The electron fluence spectrum in each voxel is computed, and dose is subsequently determined by integration over energy of the product of the electron fluence spectrum and the relevant cross section within the voxel, divided by the mass density of the voxel. 20 As such, the Acuros AXB characterizes electron transport, albeit approximately, more like a Monte Carlo-based algorithm. A GBBS algorithm such as AXB may therefore better suited to calculate dose near the various heterogeneous boundaries of the AeroForm TE.
Convolution algorithms such as Pinnacle CCC and Eclipse AAA underestimate dose perturbation occurring in the presence of highdensity materials. The most significant inaccuracies occur near the boundaries of the high-density object, where effects of electron backscatter and lateral scatter are not accounted for. AAA also underestimates attenuation from high-density materials. 12,13 For patients with TE that uses an IMP, the known inaccuracies of convolution algorithms may have a limited clinical impact. 21,22 The high-density magnet is generally more than 1 cm away from patient tissue, so inaccuracies in lateral scatter dose are confined to the saline filling the TE and the IMP itself. AAA's underestimation of attenuation through the IMP might affect the accuracy of the dose calculated in the "dose shadow" region. However, as discussed previously, this effect can be mitigated by assigning an experimentally This study aims first to describe a method for modelling the AeroForm in Eclipse TPS. It is the authors' hope that the procedure can be easily implemented into any clinician's radiation therapy TPS.
The TPS model will be optimized and evaluated for both AAA and AXB, employing techniques described Chen et al and Yoon et al.
Finally, the dosimetric impact of the AeroForm TE will be evaluated using clinical patient data. Differences in plan outcome based on dose calculation algorithm will be discussed. An understanding of dose calculation accuracy in the vicinity of the AeroForm TE is critical for assessing individual plan quality, appropriateness of different planning techniques and dose calculation algorithms, and even the decision to use the AeroForm TE in a PMRT setting.

2.A | Imaging and modeling the AeroForm TE
The AeroForm TE is available in three sizes: 400, 600, and 800 cc.
The 400 and 600 cc sizes utilize a CO 2 reservoir 7.6 cm in length and 1.9 cm in diameter. The 800 cc TE uses a reservoir 9.0 cm in length and 1.9 cm in diameter. Both sizes of AeroForm reservoir are shown in Fig. 1, with the Natrelle® IMP for comparison.
Each reservoir size was imaged using a Philips Brilliance Big Bore CT scanner (Philips Health Care, Cleveland, OH, USA). The AeroForm reservoir was aligned such that the long axis was perpendicular to axial plane, so as to minimize artifacts. Helical CTs were acquired for a 20 cm FOV with a 1 mm slice thickness using 120 kVp and 500 mAs. A 16-bit reconstruction was generated.
TPS models of both CO 2 reservoir sizes were created in Eclipse

2.B | Validation of the AeroForm TPS model
Several simple geometries were used to validate the TE models. Each expander was oriented vertically and horizontally on slabs of solid water, as shown in Fig. 4. A CT scan of each geometry was acquired and imported into Eclipse TPS. Individual components of each expander were defined by rigidly registering the acquired CT scan to the appropriate TPS model, as described previously. The solid water and air surrounding the expanders were also contoured, so appropriate densities could be assigned to correct for streaking artifacts. A single open field plan was created to deliver an AP, 20 cm 2 × 20 cm 2 , 6 MV beam to each geometry. Dose was calculated using a 1 mm grid size for each expander size and geometry, with both AAA and AXB dose calculation algorithms (version 13.5.35). In the planning CT, density overrides were applied to the air and solid water to correct for artifacts. Initially, CT values of the various components of the CO 2 reservoir were assigned based on manufacturer specifications. In AAA, initial CT values were assigned to achieve a physical density as near as possible to the vendor-stated physical density. In AXB, initial CT values were assigned based on the Eclipse default material values for stainless steel. Upon analysis, the densities of the various AeroForm components were adjusted to best fit measured data. Re-assignment of CT value, dose calculation, and comparison to measured data was then repeated iteratively to identify the optimal density assignments that resulted in the best agreement between measured and calculated data for both AAA and AXB.    Two additional structures were defined for the purpose of evaluating areas of particular interest near the CO 2 reservoir. "Chest wall AF " is defined as the tissue between the lung and expander balloon, and within 2 cm of the CO 2 reservoir. "Dose shadow AF " is a cylindrical projection of the CO 2 reservoir along its long axis in PTV_E-VAL. Both structures are shown in Fig. 5.

2.C |
The optimized CT values ( To achieve acceptable coverage, the dose shadow and chest wall regions were "boosted." An example of the resulting fluence is shown in Fig. 6.
Initial planning employed AAA dose calculation algorithm. After a satisfactory plan was achieved and reviewed by a physician, each plan was re-calculated using AXB dose calculation algorithm. Beam parameters were unchanged. Patients 1, 3, and 4 were planned to a total dose of 50 Gy, and patient 2 was planned to a total dose of

3.A | Imaging and modeling the AeroForm TE
Using optimal geometry and imaging technique, the AeroForm CO 2 reservoir alone can be exquisitely imaged without a MAR reconstruction. The CT technique described previously was used to create the TPS models of the AeroForm device in Eclipse TPS (Fig. 2). Mea- Both reconstructions were made available during contouring, but the non-MAR reconstruction was used as the primary planning CT.
Planners observed that the MAR reconstruction offered better visualization of the CO 2 reservoir and the external contour. However, near tissue-air cavity interfaces such as the chest wall-lung or PTV_EVAL-expander balloon boundaries, the MAR reconstruction sometimes generated high-density "cavity filling" artifacts, as shown in Fig. 7. Therefore, the non-MAR reconstruction was more useful in

3.B | Validation of the AeroForm TPS model
Transmission through the AeroForm CO 2 reservoir was best modelled using the RED assignments shown in Table 2. Figure 8 compares the measured and calculated dose profiles along the long axis of the AeroForm for the horizontal geometry [ Fig. 4(a)]. Calculated dose profiles are shown before and after the RED assignments were optimized, for both AAA and AXB. In AAA, the initial model (which assigned densities based on manufacturer specified physical densities) overestimated dose transmitted through both the stainless steel bottle (SSB) and the SSCW. After the RED of the SSCW was adjusted to 6.1 (   Even using the maximum RED value, transmission through the solid part of the SSB is slightly overestimated by AAA.

3.C | Evaluation of the AeroForm TPS model in clinical cases
An example planned dose distribution for Patient 3 is shown in  Tables 3 and 4, respectively. The percent difference between the AAA and AXB calculated doses are also reported in Tables 3 and 4.
The DVH plots show that, compared to AXB, AAA consistently overestimates dose in the CO 2 reservoir and in the expander balloon. D mean of the CO 2 reservoir is between 5% and 8% higher when calculated by AAA. D mean of the expander balloon is 6 to 13% higher when calculated by AAA. The difference in D max is highly variable.
The changes in dose distribution in patient tissue are more com-    to PTV_EVAL are less than 1.5% for all patients. Differences in specific subregions of PTV_EVAL appear to be somewhat patientspecific. AAA underestimated D mean to the chest wall near the CO 2 reservoir (chest wall AF ) by 1.0% to 5.0%. The shape of the DVH shoulder and tail was sharper in AAA than in AXB for patients 2, 3, and 4, whereas the DVH for patient 1 appeared to be uniformly shifted. D mean in the dose shadow region was between 0% and 5% lower when calculated using AAA compared to AXB. Changes in DVH shape did not follow a distinct pattern. This may indicate that dose distribution differences in the dose shadow and chest wall regions are anatomy-and plan-specific and cannot be easily generalized. Changes in D max were highly variable. This is partially explained by the fact that plan optimization using AAA aimed specifically to control hot spots, whereas the re-calculation with AXB did not include any re-optimization.
Differences in D max and in the shoulder and tail regions of clinically in these areas informs the optimization of the treatment plan, and thus directly affects the design of treatment beams. Beam optimization, and therefore target coverage, will be improved using the optimized TPS model of AeroForm in conjunction with a deterministic dose calculation algorithm such as AXB.

CONF LICTS OF INTEREST
The authors declare no conflict of interest.