MLC parameters from static fields to VMAT plans: an evaluation in a RT-dedicated MC environment (PRIMO)

For this work, PRIMO [14] was used from version 0.1.3.137 to 1.0.0.1756-beta following the software development updates.

PRIMO combines a graphical user interface with a general-purpose radiation transport code, PENELOPE, and the fast Dose Planning Method DPM algorithm [15, 16], specifically implemented in PRIMO for the simulation of radiotherapy beams. A complete simulation in PRIMO is divided into three segments: for the first one, S1, PRIMO allows the user to select the linac head of interest from a predefined geometry library, to tune the primary beam parameters, and perform the simulation of the upper part of the linac head. The output of S1 is a phase-space file representing the beam above the jaws. The second segment, S2, identifies the phase-space at the downstream end of the region corresponding to the bottom of the collimating devices. It includes the simulation of the secondary collimating system (both jaws and MLC). The output of S2 is a phase-space file representing the beam arranged for a specific plan, located at the bottom of the collimation system. Finally, the third segment, S3, estimates the absorbed dose within a phantom or a patient CT.

A phase-space of the S1 segment of our beam was simulated in PRIMO, using PENEASY/PENELOPE as simulation engine. For the linac head, an approximate empirical geometry named FakeBeam, developed by the PRIMO authors [17], was used with a 10 MV FFF beam. The primary beam was characterized by the following beam parameters for the initial electron beam: mean energy of 10.8 MeV, energy full–width at half–maximum (FWHM) 0, focal spot FWHM 0.1 cm, and beam divergence 0. Those are the default parameters suggested in PRIMO. The splitting roulette, a variance reduction technique described in [18] used in this work. A total number of 77 × 10⁶ histories were used for the simulation and a phase-space file (PSF) of 56 Gigabyte was obtained in segment S1. This PSF was used as the source of particles for the S2 and S3 segments, simulated together using DPM. In S2, the HD-120 MLC was selected when defining the field or importing the treatment plan (leaf geometry, to our best knowledge, is included in PRIMO, according to the manufacturers’ blueprints). During the S3 simulation, the transport parameters for the DPM included cut-off energies of 50 keV for photons and 200 keV for electrons. The S3 used, depending on the test case, water phantom, solid water phantom, or patient CT dataset. In order to reduce the statistical uncertainty, a splitting factor was applied for the S3 simulation as described in PRIMO User’s Manual [19].

PRIMO reports the average statistical uncertainty of the simulation considering all the voxels (voxel size for all the simulations was 1.5 mm in each direction) receiving more than 50% of the maximum absorbed dose and are given at 2 standard deviations. The variance reductions applied in S1 and S3 allowed obtaining uncertainties lower than 2%, except for tests with very small field sizes (5 mm), where a 3% value was accepted.

The validation of the PSF from S1 simulation was conducted against measurements, with static square fields shaped by the jaws, and not the MLC. Depth dose curves (PDD), profiles and OF at isocenter, 5 cm depth, were compared for different field sizes (2, 3, 5, 10 and 20 cm²). Measurements were acquired in a water phantom with the microdiamond detector (PTW).

The PSF obtained in the validation phase was used as the source of particles in all the simulations carried out in the tests described below

The Varian 120-HD MLC has the 32 central leaf pairs with a 2.5 mm width at the isocenter, and the remaining have a width of 5 mm, to cover a field 22 cm long. To reduce the interleaf leakage, leaf sides are designed with a ‘tongue-and-groove’ arrangement, where dovetails shape the complementary tongue or groove regions of adjacent leaves. This structure reduces the interleaf fluence when the leaf sides are exposed to the radiation beam. This fluence reduction is known as tongue-and-groove (TG) effect [20] and can lead to under-dosages [21]. All leaf ends have a rounded edge design to minimize the penumbra variation for all leaves positions.

Both the static and the dynamic behavior of the MLC were investigated in two sets of tests summarized in Table 1 and described below.

The dynamic behavior of the MLC was tested at the leaf ends with a dosimetric leaf gap (DLG), and at the leaf side with the tongue-and-groove (TG) effect. The tests were repeated with MC simulations in PRIMO, with measurements with films, and with Acuros calculations in Eclipse.

Point dose measurements (MLC_square and DLG_test) were acquired with a microDiamond detector (PTW, Freiburg, Germany, 2.2 mm radius chips) in an BluePhantom² (IBA Dosimetry) water tank.

Profile measurements (Table 1) were acquired with films in a Plastic Water phantom (MULTIcube, IBA Dosimetry).

Radiochromic EBT3 films (GafChromic, ISP Technology, Wayne, NJ) were used, calibrated with the dose-exposure curve [25]. Calibration was performed in the range 0–5 Gy (0.25 Gy spacing between 0 and 1.25 Gy, and 1 Gy between 2 and 5 Gy). The films were scanned on the green channel of a 48-bit scanner (Epson Expression 1000XL, Epson America, Sunnyvale, CA) with a resolution of 72 dpi (pixel resolution of less than 0.4 mm). The films were placed in the scanner with accurate and reproducible procedure and orientation to exclude variations in the scanner response over scan field. The calibration curve was fitted with a third grade polynomial function using the OmniPro-I’mRT software (IBA Dosimetry). The uncertainty of the film measurements in the dose range of interest in this work can be considered < 3% [26].

MLC tests were repeated on the Eclipse TPS, and calculated with Acuros dose calculation algorithm. It is a linear Boltzmann transport equation solver, expected to have similar degree of accuracy of a MC simulation.

Regarding the MLC modeling, Eclipse considers a single MLC transmission value, input by the user during the beam configuration. This neglects the transmission modifications due to energy spectrum variations in the field area, or variations between leaves of different widths, or variations of the transmission with depth. The TG is modeled separately by modifying the fluence, extending the leaf projection in the direction perpendicular to the leaf motion by a fixed parameter [24, 27]. The rounded leaf ends are modeled through the DLG as described above. This parameter is used in Eclipse modifying the fluence, as generated by shifting the leaf end position back by half of the DLG value.

The MLC parameters used in Eclipse for the Acuros configuration in this work (for the 10 MV FFF beam) were: MLC transmission of 1.3%, and DLG equal to 0.41 mm.

All the above described tests were computed with Acuros in the same conditions for subsequent comparisons, using a dose calculation grid size of 1.5 mm.

For the last phase of this work, 5 patients were selected from the institutional database, covering a wide range of target volumes (from 0.9 to 995 cm³) and plan modulations (evaluated in terms of mean segmented opening and mean segmented area) in different anatomical regions (brain, lung and breast).

The treatment plans were optimized for VMAT technique in Eclipse, using the PO (Photon Optimizer) algorithm in its version 13.5, with an optimization resolution setting of 2.5 mm. The final dose distribution was calculated with Acuros using a grid size of 1.5 mm.

Acuros calculates the energy dependent electron fluence, based on the patient material properties derived from the Hounsfield Units (HU) of the CT dataset. For each material the specific chemical elemental composition is based on the ICRP Report 23 [28] and ICRP Report 89 [29, 30].

The DICOM files (plan, structures and CT images) were exported from Eclipse and then imported in PRIMO. The dose distributions in the patients, for each plan were simulated with the DPM using a voxel size of 1.5 mm. The medium material is assigned according to the material conversion, as reported in Table 2.

The dose distributions obtained with PRIMO and Acuros, both reported as dose to medium, were compared in terms of 3D gamma analysis within the external patient contouring (3%-2 mm and 2%-2 mm) [31], using the tool implemented in PRIMO.

The simulations of static jaw-defined fields were compared with microdiamond measurements for 2, 3, 5, 10 and 20 cm² squared fields.

The average point-by-point differences between measured and calculated PDDs were < 1% for fields ≥3 × 3 cm². This result is in line with the one obtained by Hermida-López et al. [1] where the agreement between dose simulated with PRIMO and measurements was within 1.3%. Dose profiles showed average point-by-point differences below 2% for all considered field sizes. These results are in line to those obtained by Belosi et al. [6] for the Varian provided PSF for FFF beams validated with PRIMO.

Agreement between OF are within 0.4% down to the 3 × 3 cm² field. For the 2 × 2 cm² field the differences were found to be up to 1.1%. The current method to convert eV/g to Gy/MU in PRIMO does not correct for the radiation backscattered into the monitor chamber, which depends on the field size, particularly for small fields. As reported by Zavgorodni [32] the backscatter correction factor BSF, however, is small for the considered field sizes, with values of 0.2% for the 2 × 2 cm² and negligible for larger fields. Correcting the simulated output of the 2 × 2 cm² field by this BSF from a similar linac, the difference with measurements states below 1%.

The comparisons between MC and measurements showed that the PSF generated in PRIMO agrees with the 10 MV FFF beam from our EDGE linac, and further tests can be carried out. A deeper presentation of the results of the phase space validation, however, exceeds the aim of this paper.

The clinical plans were evaluated in terms of 3D global gamma index analysis (3%/2 mm and 2%/2 mm as dose difference and distance-to-agreement criteria) between MC in PRIMO and Acuros dose calculations, on the body structure (including the whole patients within the CT dataset) and the planning target volume PTV. The choice of the distance-to-agreement gamma criterion is consistent with the dose calculations resolution of 1.5 mm.

The gamma analysis resulted in an average gamma agreement index (GAI, defined as the percentage of the analyzed point passing the gamma criteria) for the body of 98.9 ± 0.6% for the 2%/2 mm criteria and 99.5 ± 0.2% for the 3%/2 mm, and for the PTV the GAI was 91.4 ± 0.6% and 97.7 ± 0.2% for the 2%/2 mm and 3%/2 mm criteria, respectively. The lowest GAI values for PTV were 88.2 and 97% for 2%/2 mm and 3%/2 mm criteria.

The difference in handling the MLC in PRIMO and in Eclipse, as described in the main part of this work, is only one of the reasons that could result in different dose estimations between MC and Acuros. Another important source of such differences is the different handling of the materials in the two systems, which can contribute in a twofold way. Firstly, the different material assignment according to the HU in Acuros and MC in PRIMO, as shown in Table 2, especially with the overlapping adjacent materials in Acuros, lead to different dose calculations due to different material assignment. For some materials it could be of few percent (e.g. ~ 2% between adipose and muscle [37]), higher when cartilage and bone structures are included. Secondly, the different elemental composition of the tissues in the two systems is not identical (as described in [37]), leading again to some differences in dose estimation.

The here presented results showed the importance of understanding and analyzing the parameters that could influence the dose calculation in the specific systems. From this work the MC management in PRIMO of the MLC presented better agreement with measurements than the beam source modeling for Acuros in Eclipse. In the common workflow of the clinical practice, the patients are treated with plans calculated by the TPS (Acuros in our work), and an independent dose calculation check is suggested to reduce errors induced by the dose calculation procedure. The interesting point in this flow is to understand the possible source of the discrepancies in order to properly judge results from the independent checks. However, we believe that the MC in PRIMO can be safely used for independent dose calculation checks, having proved its better management of MLC.