Independent calculation‐based verification of volumetric‐modulated arc therapy–stereotactic body radiotherapy plans for lung cancer

Abstract This study aimed to investigate the feasibility of independent calculation‐based verification of volumetric‐modulated arc therapy (VMAT)–stereotactic body radiotherapy (SBRT) for patients with lung cancer using a secondary treatment planning system (sTPS). In all, 50 patients with lung cancer who underwent VMAT‐SBRT between April 2018 and May 2019 were included in this study. VMAT‐SBRT plans were devised using the Collapsed‐Cone Convolution in RayStation (primary TPS: pTPS). DICOM files were transferred to Eclipse software (sTPS), which utilized the Eclipse software, and the dose distribution was then recalculated using Acuros XB. For the verification of dose distribution in homogeneous phantoms, the differences among pTPS, sTPS, and measurements were evaluated using passing rates of a dose difference of 5% (DD5%) and gamma index of 3%/2 mm (γ3%/2 mm). The ArcCHECK cylindrical diode array was used for measurements. For independent verification of dose‐volume parameters per the patient’s geometry, dose‐volume indices for the planning target volume (PTV) including D95% and the isocenter dose were evaluated. The mean differences (± standard deviations) between the pTPS and sTPS were then calculated. The gamma passing rates of DD5% and γ3%/2 mm criteria were 99.2 ± 2.4% and 98.6 ± 3.2% for pTPS vs. sTPS, 92.9 ± 4.0% and 94.1 ± 3.3% for pTPS vs. measurement, and 93.0 ± 4.4% and 94.3 ± 4.1% for sTPS vs. measurement, respectively. The differences between pTPS and sTPS for the PTVs of D95% and the isocenter dose were −3.1 ± 2.0% and −2.3 ± 1.8%, respectively. Our investigation of VMAT‐SBRT plans for lung cancer revealed that independent calculation‐based verification is a time‐efficient method for patient‐specific quality assurance.


| INTRODUCTION
Stereotactic body radiotherapy (SBRT) is an effective treatment for patients with inoperable stage I non-small-cell lung cancer (NSCLC). 1 Previous studies found that SBRT for NSCLC can achieve excellent 3-year overall survival (OS) and high local control rates with minimal toxicity. [2][3][4][5] In recent years, volumetric-modulated arc therapy (VMAT) has been introduced as a form of SBRT for lung cancer treatment 6 and it is becoming regularly employed owing to its rapid delivery of radiation doses and superior dose conformity. 7 Since April 2018, VMAT has been used in our institution to treat patients with NSCLC using 6 MV X-rays delivered by the Vero4DRT linear accelerator system (Mitsubishi Heavy Industries, Ltd., Tokyo, Japan and Brainlab AG, Feldkirchen, Germany).
Patient-specific quality assurance (QA) should be performed for all radiotherapy plans. For three-dimensional conformal radiotherapy (3D-CRT)-type SBRT for NSCLC, we employed an independent calculation-based verification using a secondary treatment planning system (sTPS). 8 The dose difference to the planning target volume (PTV) isocenters delivered by the X-ray voxel Monte Carlo (XVMC) versus the Acuros XB (AXB) at our institution was found to be −0.3 ± 1.4% (the XVMC and AXB are categorized as type "c" algorithms in which the modeling of secondary electron transport is markedly improved compared to superposition/convolution methods), and we defined the tolerance level as the isocenter dose difference. While measurementbased patient-specific QA methods are still widely used for IMRT/ VMAT as recommended by the "American Association of Physicists in Medicine" task group (218), 9 independent calculation-based verification was one of the patient-specific QA methods used for the non-intensity modulated radiotherapy (IMRT) technique. 10 As such, measurement-based patient-specific QA for VMAT-SBRT is performed at our institution using an ArcCHECK cylindrical diode array (SunNuclear, Melbourne, FL, USA). Although the actual treatment delivery time is shorter owing to VMAT specifications, performing patient-specific QA consumes more time than does the 3D-CRT itself.
Recent studies in the medical physics field have explored the efficiency of non-measurement-based patient-specific QA for IMRT-VMAT. 11,12 Tachibana et al. 13 compared secondary checks of 973 treatment planning protocols using computer-based independent verification for non-IMRT, IMRT, and VMAT methods, and found that a 5% action level was justifiable for all sites except the lungs. Incidentally, the dose calculation algorithms in their study included the pencil beam method, which is not suitable for lung-related calculations. Thus, there were large systematic differences in lung site estimates when using computer-based independent verification because of the large differences in heterogeneity corrections between the primary treatment planning system (pTPS) and verification program. Handsfield et al. found that a new patient-specific QA procedure for TomoTherapy using log files and secondary Monte Carlo dose calculations was an effective and efficient alternative to the traditional phantom-based QA method. 14 However, their analyses required commercially available or special in-house software. Considering the necessity of patient-specific QA for IMRT-VMAT, including for pulmonary and non-pulmonary sites, we turned our attention to an sTPS for which commercially available software is used in our facility.
This study aimed to investigate the feasibility of using a commercially available software, Eclipse, as an independent calculation-based verification for lung VMAT-SBRT. Specifically, dose distributions in a homogeneous phantom and patient geometry were calculated and measured. Such a verification system would be more efficient for evaluating dose distributions in pulmonary sites than computer-based verification.  15 The range of lung tumor motion, which was evaluated as the mean ± standard deviation (SD) (minimu to-maximum), was 6.0 ± 4.1 (0-20) mm in the superior-inferior direction, 2.5 ± 1.3 (0-6) mm in the left-right direction, and 3.0 ± 1.7 (0-10) mm in the anterior-posterior direction. The maximum and mean intensity projection images were acquired. The dose calculation was performed on the mean intensity projection images.

2.B | Target delineation and fraction regimens
The internal gross tumor volume (iGTV) was delineated based on the maximum intensity projection as well as the 10 respiratory phase  Patients who underwent VMAT-SBRT were treated with the Ver-o4DRT system; all VMAT-SBRT plans were created using RayStation version 6.2 (RaySearch Medical Laboratories AB, Stockholm, Sweden), which we considered the pTPS. Collapsed-Cone Convolution (CCC), version 3.4, was used as the dose calculation algorithm, and the dose calculation grid size was 2.0 mm. All the plans were optimized using a gantry angle sampling of 4°between the control points. VMAT-SBRT was delivered with 2-6 arcs, including coplanar and non-coplanar beams.

2.C.2 | sTPS
After the VMAT-SBRT plans were created using the pTPS, all DICOM files (including the CT images, structure files, plan files, and dose files) were transferred from the pTPS to the sTPS, which was Eclipse version 15.6 (Varian Medical Systems, Palo Alto, CA). The dose calculation algorithm was AcurosXB (AXB) version 15.6.05, and the dose calculation grid size was 2.0 mm using dose-to-medium with heterogeneity correction. Details of the commissioning of AXB for Vero4DRT including dosimetric evaluation for a heterogeneity phantom were described previously. 8,16 The dose comparison between pTPS and sTPS was performed in Eclipse.

2.D | Verifications
The following verifications were performed for 50 treatment plans of VMAT-SBRT: 1. Independent verification based on dose distributions in homogeneous phantoms.
2. Independent verification based on dose-volume parameters using patient geometry.

2.D.1 | Independent verification based on dose distributions in the homogeneous phantom
For this verification, two comparisons were performed. The first was a comparison between measured and calculated dose distributions, including the pTPS and sTPS, as described in "C. Patient-specific QA" section. Therefore, the dose difference represented the uncertainty in the treatment plan, including the linear accelerator output variations, multileaf collimator position accuracy, or the TPS beam modeling accuracy. The second one was a comparison between the dose distributions for pTPS and sTPS; the dose differences mainly represented the error in the TPS's model except for the effect of heterogeneity on patient geometry. These differences were evaluated using DD3%, DD5%, γ2%/2 mm, γ3%/2 mm, and γ3%/3 mm.
In addition to the dose index verifications, the similarity of fail points among the pTPS, sTPS, and measurement dose distributions were evaluated by Simpson's Faunal Resemblance Index (FRI). 17 The FRI is used to calculate the similarity between pairs of community samples. The FRI takes into account only the number of species occurring in the smaller sample; thus, it is the least influenced by the sample size and emphasizes the similarity of fail points.
The formula used for FRI was as follows: where |X ∩ Y| is the number of non-empty categories in the inter-  between the sTPS and pTPS, defined as follows:   Table 1 shows the dose indices for the dose distributions per pTPS, sTPS, and the actual measurement for 50 patients. The mean dose index derived from pattern 3 was slightly better than that derived from pattern 2. On the other hand, the mean dose index of pattern 1 was higher than those of patterns 2 and 3. The uncertainty of the pattern 1 treatment plan was larger than the error in the TPS's calculation. Figure 2 shows the examples of FRI region comparisons between patterns 1 and 2 and between patterns 1 and 3 in terms of the DD3% and DD5%. In this example, the min (|X|, |Y|) represented the total fail points in pattern 1 because this number was smaller than that of the total fail points in patterns 2 or 3. In the FRI region of DD3%, the same fail points were observed in patterns 1 and 2 and for patterns 1 and 3. In the FRI region of DD5%, patterns 1 and 3 shared a common region although patterns 1 and 2 did not. Table II shows the FRIs between patterns 1 and 2 as well as those between patterns 1 and 3 in terms of the DD3% and DD5%. The FRIs were calculated by excluding the 100% passing rates for DD3% and DD5% (as these contained no fail points). For both DD3% and DD5%, the FRIs between patterns 1 and 3 were higher than those between patterns 1 and 2, especially for the DD5%.   Table 4. Doses to the isocenter and target volume per the sTPS were significantly smaller than those per the pTPS. On the other hand, the dose to the lung per the sTPS was larger than that per the pTPS; this difference was significant except for the V 20Gy of the lung. however, this study reported no details regarding the calculation algorithm. They concluded that the Eclipse-based sTPS was an accurate, robust, and time-efficient method for patient-specific IMRT QA.

3.B | Dose-volume parameters in patient geometry
In our study, we also found that the dose indices obtained using AXB were better than those of the CCC when comparing each of these to the actual measurement. Moreover, our study is the first one to evaluate the fail points among dose distributions of pTPS, sTPS, and actual measurement. In general, the sTPS using the AXB had good agreement with the measured dose distribution; this agreement was better than that of the pTPS using the CCC. 19 Thus, the fail points derived from the sTPS vs. actual measurement ought to reflect the uncertainty of the treatment plan more reliably than those derived from the pTPS vs. measurement. We also found that the FRI for the DD5% for patterns 1 vs. 3 was 0.48 ± 0.41. In other words, most of the fail points in relative to DD5% are related to the uncertainty in the treatment planning dose calculation algorithms.
Therefore, when considering calculation-based verification without actual measurements, evaluating the gamma passing rate between pTPS and sTPS is useful to determine treatment plan uncertainty. In addition, we found that the FRI between patterns 1 and 2 as well as those between patterns 1 and 3 were lower than 1. This was because the calculation and measurement uncertainties had different factors, for example, beam modeling accuracy or measurement device inaccuracies. Un-passing points of FRI means different cause T A B L E I Dose indices for pTPS, sTPS, and measurement dose distributions for 50 treatment plans. were useful for independent verification. In our study presented here, we also found that the dose differences and deviations were the smallest at the isocenter.
The dose differences among dose calculation algorithms are large for non-homogeneous regions such as the lung. Tsuruta et al. reported significant differences between dose calculation algorithms around the F I G . 2. Example of the FRI regions between patterns 1 and 2 for DD3% (a) and DD5% (b) as well as between patterns 1 and 3 for DD3% (c) and DD5% (d). Pattern 1, 2, and 3 were defined between pTPS and sTPS, pTPS and measurement, and sTPS and measurement, respectively.
T A B L E I I FRIs between patterns 1 and 2 and between patterns 1 and 3 in terms of the DD3% and DD5% for 50 treatment plans. PTV in low-density regions when performing dosimetric comparisons of the AAA, AXB, and XVMC for lung cancer. 16 In particular, the dose difference between the AAA and XVMC outside the PTV was up to 15.5%. As shown in [ Fig. 3 | 141

| CONCLUSIONS
The findings of this study were as follows: (a) calculating the difference in the gamma passing rates of pTPS and sTPS is useful for determining treatment plan uncertainty, (b) small deviations in the DEs of target dose-volume parameters in pulmonary sites are acceptable, and (c) isocenter dose verification is suitable for defining the tolerance level for patient-specific QA. Independent calculationbased verification can be used as a time-efficient method for patient-specific QA under the condition that pre-treatment verification is performed to confirm the data transfer.

ACKNOWLEDG MENTS
We acknowledge financial support by a Grant-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science, and Technology of Japan (Grant No. 18K15545).

CONFLI CT OF INTEREST
The authors have no relevant conflicts of interest to disclose.
T A B L E I V DE and VE of dose-volume parameters of the isocenter, iGTV, PTV, and the lung and spinal cord for 50 treatment plans.