A simple technique to improve calculated skin dose accuracy in a commercial treatment planning system

Abstract Radiation dermatitis during radiotherapy is correlated with skin dose and is a common clinical problem for head and neck and thoracic cancer patients. Therefore, accurate prediction of skin dose during treatment planning is clinically important. The objective of this study is to evaluate the accuracy of skin dose calculated by a commercial treatment planning system (TPS). We evaluated the accuracy of skin dose calculations by the anisotropic analytical algorithm (AAA) implemented in Varian Eclipse (V.11) system. Skin dose is calculated as mean dose to a contoured structure of 0.5 cm thickness from the surface. The EGSnrc Monte Carlo (MC) simulations are utilized for the evaluation. The 6, 10 and 15 MV photon beams investigated are from a Varian TrueBeam linear accelerator. The accuracy of the MC dose calculations was validated by phantom measurements with optically stimulated luminescence detectors. The calculation accuracy of patient skin doses is studied by using CT based radiotherapy treatment plans including 3D conformal, static gantry IMRT, and VMAT treatment techniques. Results show the Varian Eclipse system underestimates skin doses by up to 14% of prescription dose for the patients studied when external body contour starts at the patient's skin. The external body contour is used in a treatment planning system to calculate dose distributions. The calculation accuracy of skin dose with Eclipse can be considerably improved to within 4% of target dose by extending the external body contour by 1 to 2 cm from the patient's skin. Dose delivered to deeper target volumes or organs at risk are not affected. Although Eclipse treatment planning system has its limitations in predicting patient skin dose, this study shows the calculation accuracy can be considerably improved to an acceptable level by extending the external body contour without affecting the dose calculation accuracy to the treatment target and internal organs at risk. This is achieved by moving the calculation entry point away from the skin.


| INTRODUCTION
Skin dose and its resultant toxicity, radiation dermatitis, has long been a concern of the radiation oncologist and is often a dose limiting toxicity of high-dose treatments, particularly in head-and-neck and thoracic cancer patients. 1-5 Improvements in radiation therapy treatment technique and immobilization devices reduce patient setup uncertainty but have exacerbated this clinical dilemma. It is well known that immobilization devices have a noteworthy deleterious effect on patient skin dose. 6-9 Therefore, accuracy of predicting skin dose by commercial treatment planning systems (TPS) is critical, as skin dose toxicity has a major impact on how well a patient tolerates treatment. It is known that model based dose calculation algorithms have limitations at the buildup region where the charge particle equilibrium (CPE) is not established. Therefore, accurate skin dose calculations would greatly help clinicians make appropriate treatment plan decisions for these patients where skin toxicity has historically been an issue. 10 A number of investigations have focused on the accuracy of skin dose or entrance dose (or surface dose as sometimes referred) calculations in commercial TPS. [11][12][13][14][15][16][17] Court et al. 11 studied pencil beam convolution (PBC) algorithms in the Varian (Varian Medical Systems, Palo Alto, CA, USA) Eclipse system by comparing measured entrance dose to the Eclipse predicted dose. Oinam and Singh 15 measured entrance dose in a phantom for a seven field 6 MV energy IMRT case and compared to the calculated dose by PBC algorithm and anisotropic analytical algorithm (AAA) in Eclipse version 8.6. Their results showed that AAA was more accurate than PBC, but both have limitations on predicting entrance dose accurately at depth less than 0.2 cm. Panettieri et al. 12  The heterogeneity correction is employed in all AAA dose calculations; and, unless otherwise specified, the calculation grid size is 2.5 mm, which is a typical size in clinical practice. Although the grid size may affect the skin dose calculation accuracy, switching from 2.5 mm grid size to 1 mm grid size, which is smallest for AAA in Eclipse, only slightly improves the accuracy. 20 In addition, the calculation time is much longer with 1 mm grid size, making it clinically unattractive. Furthermore, the grid size for MC dose calculations in this study (see below) is also set at 2.5 mm, which makes the comparison meaningful and justified. To avoid confusion, the skin dose is defined in this study as the mean dose to the skin structure of 5 mm thickness for the CT based dose calculations. To quantify the skin dose, the skin was contoured to be an area of 2 9 2 cm 2 , corresponding to a volume of about 2 cm 3 , which is of clinical interest.
The skin dose predicted by Eclipse is compared with that of MC calculations which are benchmarked by measurements in phantoms.
The term "entrance dose" is used for the phantom measurements.

2.B | Monte Carlo simulations
The MC simulation code used in this study is the EGSnrc 21 code and its user codes BEAMnrc 22,23 and DOSXYZnrc. 24 The modulated realistic beams from the Varian TrueBeam accelerator with a Millennium 120 multileaf collimator (MLC) have been simulated by using BEAMnrc/DOSXYZnrc codes and calculated dose distributions have been validated. [25][26][27] Varian TrueBeam phase-space files 28 (version 2.0) are used as the radiation sources at the plane before entering the secondary collimators, or jaws, and MLC. The jaw openings and MLC modulations are modeled in detail in the simulations as described in the study by Lobo and Popescu. 27 The typical source phase-space file for each beam energy used for simulation is about 20 GB in size containing about 900 million particles. The large number of particles used is necessary to achieve a statistical uncertainty of about 1% for MC calculations with a calculation grid size of 2.5 mm, which is the same as in Eclipse calculations. The EXACT boundary crossing algorithm is used and the electron cutoff energy (ECUT) is set at 0.7 MeV.

2.C | Measurements
MC calculations have to be validated by measurements if they are used as a benchmarking tool. The EGSnrc MC code used in this study has been validated before on entrance dose calculations. 2,29 A further experimental validation was performed in this study where the MC calculated entrance dose was compared to that of phantom measurements using the nanoDot TM detector (Landauer Inc., Glenwood, IL, USA), the optically stimulated luminescence (OSL) dosimeters, which were pre-screened and have a measurement uncertainty of 3-5% in this study. A well-controlled geometry in which four solid water slabs of size 30 9 30 9 5 cm 3 were stacked together to form a 30 9 30 9 20 cm 3 phantom validated the accuracy of MC simulations. Photon beams of energies 6, 10, and 15 MV from a TrueBeam unit were delivered from lateral side at a gantry angle of 90 degrees, with a beam size of 4 9 4 cm 2 and 90 cm source-to-surface F I G . 1. Comparisons of surface doses (entrance and exit) measured by OSL dosimeters and the doses calculated by Monte Carlo (MC) for (a) 6 MV, and (b) 15 MV photon beams of field size 4 9 4 cm 2 , delivered 200 MU on a water slab of 20 cm thickness. The arrows indicate the in-air data points for which MC calculations include simulation of OSL dosimeter of size 10 9 10 9 1.5 mm 3 . The OSL measurement uncertainty is 3-5%, and the MC calculation statistical uncertainty is 1%. Also shown are schematic diagram and the photo of experimental arrangement. One OSL detector is on the thin tape in air and the other is on the solid water as shown in the photo insert. distance (SSD). This field size is selected as an example where the field size is big enough to establish lateral charged particle equilibrium. OSL dosimeters were placed at the central axis (CAX) on both proximal and distal phantom surfaces so that both the entrance and exit dose could be measured. The dosimeters were also placed at CAX in air at 1.2 cm distance away from the phantom surfaces to get the in-air readings. In the MC calculations, air was filled outside the solid water phantom. Three patient treatment plans were investigated which included different photon beam energy and treatment sites (H&N, lung, etc.).
The treatment techniques that utilized 3D and VMAT were included in the evaluation.
The first case is a head-and-neck (H&N) cancer patient treated with 6 MV VMAT full arcs (Fig. 2). The default patient external body contour is constructed after CT images are imported into Eclipse.
Dose calculations are performed only within this body contour, and no dose deposition outside of body contour [ Fig. 2(a)]. In reality, there is always air present outside the patient's body, due to patient supporting/immobilization devices, such as a thermoplastic head mask. Immobilization devices can cause noteworthy effect on patient's skin doses, 6, 8,9 and to quantify the effect, the external body contour is extended to include the head mask and couch table [ Fig. 2(b)]. Since it would be tedious task to make a detailed head mask contour, we simply extended a 2 cm air layer to the default body contour in addition to including the couch top within the body contour [ Fig. 2(c)]. This can be done very easily with the margin tool In case #2 the target is a right lung tumor treated with 6 MV VMAT partial arcs. Two skin structures, anterior skin and right skin, are contoured at the respective locations of the body. In case #3 the treatment target is a rib treated with AP/PA opposing beams with 15 MV photon beams. One skin structure is contoured on the anterior chest. F I G . 5. Dose difference between AAA and MC, with and without the air gap, from skin surface to 2 cm depth. (a) for the H&N dose profiles in Fig. 3(b); (b) for the H&N dose profiles in Fig. 4(b).