Development of a Geant4‐based independent patient dose validation system with an elaborate multileaf collimator simulation model

Abstract Despite the improvements in the dose calculation models of the commercial treatment planning systems (TPS), their ability to accurately predict patient dose is still limited. One of the limitations is caused by the simplified model of the multileaf collimator (MLC). The aim of this study was to develop a Monte Carlo (MC) method‐based independent patient dose validation system with an elaborate MLC model for more accurate dose evaluation. Varian Clinac 2300 IX was simulated using Geant4 toolkits, after which MC commissioning with measurements was performed to validate the simulation model. A DICOM‐RT interface was developed to obtain the beam delivery conditions including the hundreds of MLC motions. Finally, the TPS dose distributions were compared with the MC dose distributions for water phantom cases and a patient case. Our results show that the TPS overestimated the absolute abutting leakage dose in the closed MLC field, with about 20% more of the maximum dose than that of the MC calculation. For water phantom cases, the dose distributions inside the target region were almost identical with the dose difference of less than 2%, while the dose near the edge of the target shows difference about 10% between Geant4 and TPS due to geometrical differences in MLC model. For the patient analysis, the Geant4 and TPS doses of all organs were matched well within 1.4% of the prescribed dose. However, for organs located in areas with high ratio of leaf pairs with distances less than 10 mm leaf pair (LP (<10mm)), the maximum dose of TPS was overestimated by about 3% of the prescribed dose. These dose comparison results demonstrate that our system for calculating the patient dose is quite accurate. Furthermore, if the MLC sequences in treatment plan have a large ratio of LP (short), more than 3% dose difference in normal tissue could be seen.

pendent patient dose validation system with an elaborate MLC model for more accurate dose evaluation. Varian Clinac 2300 IX was simulated using Geant4 toolkits, after which MC commissioning with measurements was performed to validate the simulation model. A DICOM-RT interface was developed to obtain the beam delivery conditions including the hundreds of MLC motions. Finally, the TPS dose distributions were compared with the MC dose distributions for water phantom cases and a patient case.
Our results show that the TPS overestimated the absolute abutting leakage dose in the closed MLC field, with about 20% more of the maximum dose than that of the MC calculation. For water phantom cases, the dose distributions inside the target region were almost identical with the dose difference of less than 2%, while the dose near the edge of the target shows difference about 10% between Geant4 and TPS due to geometrical differences in MLC model. For the patient analysis, the Geant4 and TPS doses of all organs were matched well within 1.4% of the prescribed dose. However, for organs located in areas with high ratio of leaf pairs with distances less than 10 mm leaf pair (LP (<10mm) ), the maximum dose of TPS was overestimated by about 3% of the prescribed dose. These dose comparison results demonstrate that our system for calculating the patient dose is quite accurate. Furthermore, if the MLC sequences in treatment plan have a large ratio of LP (short) , more than 3% dose difference in normal tissue could be seen.  4 This shifting distance is called as a dosimetric leaf gap (DLG) and many trials have endeavored to find the optimal DLG, with the goal of minimizing uncertainty in the typical patient plan. [5][6][7][8][9] However, variations in leaf end shape cause the dosimetric effect to vary due to the irregular shape and size of the resulting fields; therefore, each dosimetric effect should be verified individually. 7 In the case of dynamic MLC, the dosimetric effect of the radiation transmitted and scattered from the rounded leaf ends can exceed 10% of the total dose. [10][11][12] Even a 1% improvement in dose delivery precision has been reported to increase the cure rate for early stage tumors by 2%. 13 Moreover, a 5% change in dose can result in 10-20% change in tumor control probability or up to 20-30% change in normal tissue complication rates if the prescribed dose falls along the steepest region of the dose-effect curve. 14 According to the guidelines in the AAPM Task Group 53 (1998) and 119 (2009) publications, two-dimensional (2D) planar dosimetry measurements (e.g., film) are recommended for evaluating the accuracy of TPS. However, 2D planar dosimetry measurements are limited in that they can only detect inaccuracies within the selected plane of treatment volume or organ at risk. 6,15,16 While three-dimensional (3D) dosimetry techniques are available, these techniques require multiple large-volume detectors such as a radiochromic plastic dosimeter and a polymer gel, in addition to scanners for verifying 3D dose distribution such as an MRI machine and an optical-computed tomography (CT) scanner. 6,17,18 The Monte Carlo (MC) method is considered the "gold standard" in assessing dose distribution and is one of the most appropriate methods for overcoming these limitations. This method has been applied to validate patient-specific IMRT dose. [19][20][21][22][23][24][25][26][27][28][29] EGSnrc/ BEAMnrc is an optimal and efficient MC code for simulating linear accelerators (linacs) whose accuracy has been validated through many dosimetry studies. 30 Accordingly, most studies aiming to develop a Monte Carlo-based radiotherapy planning system are based on EGSnrc/BEAMnrc. However, DYNVMLC, which can model the geometry of the 120-leaf Varian Millennium MLC in the BEAMnrc code, is designed to model the leaf end as a simple round shape. 31 Since the actual shape of the leaf end of the Varian Millennium MLC is originally consisted of a circular arc at the center [ Fig. 1(d)], two flat regions and two circular arcs with different radii and inner angle, dose evaluation error may occur due to incorrect leaf shape assumptions. In contrast, Geant4 is capable of sophisticated modeling of complex structures and is able to computational human phantoms and simulation of DNA strand damage. 32,33 In particular, Geant4 is able to handle dynamic geometry changes, which significantly facilitates true four-dimensional Monte Carlo simulations, for example, dynamic MLC motion, patient organs, rotating machine parts, and moving scanners. 34,35 Although the calculation time of Geant4 is longer than that of EGSnrc/BEAMnrc, this obstacle could be overcome by increasing the computational power and implementing multithreading features.
The aim of this study was to develop an independent dose calcula- The rounded edge consisted of four solid shapes such as two quarter-circles, one subtracted sector of a circle, two trapezoids, and/or one box. These solids were used to design a complex geometrical figure. In contrast, the body section was modeled simply by the G4ExtrudedSolid class. Ultimately, these solids were independently positioned in one mother volume. The Varian Millennium MLC consists of 60 leaf pairs (LPs) which have three types of leaf: full, half, and outboard. Each leaf design is mirrored on the opposite side of the bank. We defined x mm distance between paired opposite leaf tips as LP (x mm) in this study. Systems, Palo Alto, CA) to tune the characteristics of the initial electron beam. [36][37][38][39] The dimensions of the water phantom (e.g., blue phantom 2 scanning volume) were 48 × 48 × 41 cm 3 and the PDD and lateral dose distributions were calculated with 2 × 2 × 2 mm 3 voxels. and 1.00 × 1.00 × 1.00 mm 3 voxels were used for the Geant4 and TPS, respectively.

2.C | Development of the DICOM-RT interface
For MC simulations of radiotherapy plans, the automated DICOM-RT interface is essential due to many beam delivery parameters in the treatment plans. The DICOM files for radiotherapy planning consist of four file types: CT images, RT structures, RT plans, and RT doses.
These files have a format for storing information associated with a value representation (VR) that indicates the encoding type and a tag that uses 8-digit hexadecimal numbers. 40 To extract patient-dependent parameters, data can be discriminated using the DICOM tag while  Kim et al. 41 reported that Geant4-based MC dose distributions can be significantly affected by the material conversion method. Therefore, the Schneider material conversion method was used for patient cases. 42 If a special volume (e.g., the fiducial marker, virtual water phantom, and couch) is present in the CT image, EC and MD would be defined in the region of interest (ROI) based on the physical property value stored in RT structure file; this process was automated in our system.

3.B | Experimental validation of the MLC simulation model
The second validation process of the in-house system was film measurement to validate the geometrical modeling of MLC. Figure 5 shows  region differ between the Geant4 and the TPS. The dose profile of (a) in Fig. 6 shows about 0.1 Gy difference (about 16% of the maximum dose) at the dose fall-off region, this difference could be due to the assumption of the DLG value in TPS. Furthermore, the profiles (b) and (c) indicate that the TPS overestimates the absolute dose of abutting leakage, up to 20% of the maximum dose than the dose assessed by the Geant4. As discussed in Fig. 5, the overestimation of the dose of the TPS in the outer penumbra region was also observed in Fig. 6 and it was about 1.5% of the maximum dose. This means that if the ratio of the LPs (short) is dominant in the complex VMAT/IMRT plan, the dose in the normal organs or tissues could be overestimated. Moreover, the dose difference map in Fig. 6 shows that the dose at the edges of the field in the y-direction differs by about 7% of the maximum dose.   models of tongue-and-groove for shielding interleaf leakage radiations could be a reason for the different slope angles.
As found in the validation process of MLC simulation model, more than 10% dose difference was found in some local areas of VMAT/IMRT dose distribution. The dose difference could be due to the difference in the dose grid size that is quite sensitive to the LPs (short) of the MLC movement plan. Figure 9 shows  | 103 yellow patterns ranging from 0 to 3% dose difference and these patterns could be caused by the absolute dose difference resulting from LPs (<10mm) . The dose differences at the region indicated by the purple arrows are over 7%. We assumed that the high dose difference only near the surface of the water phantom could be caused by the difference in the dose grid sizes which for the Geant4 and the TPS were 1.17 and 1.00 mm, respectively. As the depth of the water phantom becomes deeper, it is presumed that the dose difference is reduced due to the phantom scatter. However, despite these local dose differences, the dose comparison study in the water phantom indicates almost identical dose distributions between the Geant4 and TPS.
3.C.2 | Patient case Figure 10 and Table 1  organs, respectively. The dose differences in Table 1 were calculated based on the prescribed dose.  (Table 1).
The patient dose distributions and the two leaf-ends distance maps for the patient case in which the two beams are rotated 360 degrees in opposite directions are illustrated in Fig. 11. Beam 2 has a larger ratio of LPs (short) than beam 1. In the map of beam 2, the ratio of LPs (short) at control points between 1 and 80 is higher than that from 81 to 178. Since the gantry rotates about 2 degrees for a control point, the total rotation angle can be assumed as 160 degrees from the first to the 80th control point. In the dose difference maps, green and blue contours indicate higher dose of TPS than that of Geant4 and especially the higher dose of TPS is noticed with the fan-shaped distribution (pink dotted line) on the right side of the XY-plane. We observed a relationship between the higher dose distributions of TPS and the distributions of LPs (short) . Because the pancreas was placed in the region of higher TPS dose, there was 3% difference of D 2% . Moreover, the differences of four PTV dosevolumetric parameters between Geant4 and TPS were about 1% or less, whereas the relative differences of the GTV (smaller volume than PTV) parameters were as high as 2.2%. We assumed that if a volume of organ or tissue of interest is very small, the dose-volumetric parameters of corresponding organ or tissue could be sensitive to the local dose differences caused by the LPs (short) .

| CONCLUSION
In this study we developed a Geant4-based independent patient dose validation system including a finely modeled MLC and automated DICOM-RT interface. The developed system was validated by three processes: MC commissioning of the modeled linac, experimental validation of the modeled MLC, and dose comparison in water between the commercial TPS and Geant4. Finally, the patient dose distribution calculated by the TPS for an abdomen case of the VMAT plan was evaluated using developed MC system.
As a result of the validation, it was confirmed that the in-house MC system was able to accurately evaluate the patient dose sufficiently. However, we found that the rounded leaf end of MLC could cause the dose difference compared to the TPS in the case of LPs (<10mm) . Da Rosa et al. 45 investigated the influence of lung heterogeneity on dose distribution in a soft tissue phantom. They evaluated PDD curves in the phantom by comparing between the dose calculated by MC method, by TPS with four algorithms, and experimental data according to the different field size from 1 × 1 cm 2 to 10 × 10 cm 2 . 44 As the results of this study, the dose difference was increased up to about 40% in the region of lung-tissue equivalent material comparing between MC and AAA for the 1 × 1 cm 2 field due to the lateral electronic disequilibrium effect. 44 In our results, about 3% difference of the prescribed dose in the normal tissue could cause by a large ratio of LPs (<10mm) in the treatment plan, even though the patient case is not that heterogeneous case. In other words, the effect of the short leaf-ends distance in a highly heterogeneous region can result in a significant dose difference. Therefore, it is necessary to quantitatively analyze the correlation between the ratio of LPs (short) and the dose difference. In the future, several treatment plans for highly heterogeneous media (e.g., lung case, head & neck case, and dummy shield case) will be evaluated with the currently developed dose validation system.

CONFLI CT OF INTEREST
No conflicts of interest.