Trajectory log analysis and cone‐beam CT‐based daily dose calculation to investigate the dosimetric accuracy of intensity‐modulated radiotherapy for gynecologic cancer

Abstract This study evaluated unexpected dosimetric errors caused by machine control accuracy, patient setup errors, and patient weight changes/internal organ deformations. Trajectory log files for 13 gynecologic plans with seven‐ or nine‐beam dynamic multileaf collimator (MLC) intensity‐modulated radiation therapy (IMRT), and differences between expected and actual MLC positions and MUs were evaluated. Effects of patient setup errors on dosimetry were estimated by in‐house software. To simulate residual patient setup errors after image‐guided patient repositioning, planned dose distributions were recalculated (blurred dose) after the positions were randomly moved in three dimensions 0–2 mm (translation) and 0°–2° (rotation) 28 times per patient. Differences between planned and blurred doses in the clinical target volume (CTV) D98% and D2% were evaluated. Daily delivered doses were calculated from cone‐beam computed tomography by the Hounsfield unit‐to‐density conversion method. Fractional and accumulated dose differences between original plans and actual delivery were evaluated by CTV D98% and D2%. The significance of accumulated doses was tested by the paired t test. Trajectory log file analysis showed that MLC positional errors were −0.01 ± 0.02 mm and MU delivery errors were 0.10 ± 0.10 MU. Differences in CTV D98% and D2% were <0.5% for simulated patient setup errors. Differences in CTV D98% and D2% were 2.4% or less between the fractional planned and delivered doses, but were 1.7% or less for the accumulated dose. Dosimetric errors were primarily caused by patient weight changes and internal organ deformation in gynecologic radiation therapy.

CBCT images improves the repeatability of patient positioning as compared with conventional two-dimensional matching. 2 However, the matching of planning CT and CBCT images is performed on the assumption that there are no anatomical changes. In general, patient weight changes, tumor shrinkage, and deformation of the internal organs occur during the treatment period over 1 month. In particular, the daily variations of bladder and rectum filling must be considered for radiation therapy of the pelvic region.
Many studies have proposed methods for evaluation of the dosimetric effect of anatomical changes on the dose of the treatment day by using the deformable image registration (DIR) technique. [3][4][5][6] Although the DIR technique enables assessment of the dosimetric effect of internal organ deformations, variations in patient weight are not considered when assessing the effect on the dose of the day. 7 In addition, low contrast and artifacts of CBCT images cause inaccuracies in DIR. 8 In this study, we investigated dose calculations by using CBCT images to evaluate the dosimetric effect of three factors: mechanical control accuracy of the treatment machine, patient setup errors, and interfractional geometric variations of target and other structures. We classified the factors related to dosimetric differences between planned doses and delivered doses into three categories and quantitatively evaluated their effect on delivered doses.

2.A | Patient population
Thirteen consecutive patients who underwent IMRT of the whole pelvis to treat cervical carcinoma at our institution were selected.
This retrospective study was approved by the ethics committee of our institution.

2.B | Contouring
To improve the precision of the patient positioning reproducibility, knees and ankles were fixed with a Vac-Lok positioning bag (CIVCO, Kalona, IA) for all patients. Planning computed tomography (CT) images were acquired with a 3-mm slice thickness (Toshiba Aquilion LB, Canon Medical Systems, Ōtawara, Japan). All patients were instructed to empty their bladders and rectums, but two were instructed to drink 200 mL of water 1 hour before CT scans and each treatment to spare the small bowel.
The clinical target volume (CTV) was defined as all areas of primary tumor and regional lymph nodes by an experienced radiation oncologist. The delineations of pelvic lymph nodes were followed by the guidelines of the Japan Clinical Oncology Group Gynecologic Cancer Study Group. 9 The internal target volume of the cervix was defined as the volume expanded 15 mm in all directions, excluding bones. Then, the planning target volume (PTV) margin of 7 mm in all directions around the CTV was added to take into account setup errors and uncertainties of inter/intrafractional organ motions. The details of contouring CTVs and organs at risk (OARs) were reported in Ref. [10].

2.C | Treatment planning
All patients were planned by using an Eclipse version 13.6 (Varian Medical Systems, Palo Alto, CA) and an analytical anisotropic algorithm. The resolution used for dose calculation was 2 mm in all directions. IMRT plans were optimized on a seven-or nine-beam dynamic multileaf collimator (DMLC) delivery system using a 10-MV photon beam produced by a TrueBeam linear accelerator with a Millenium 120 MLC (Varian Medical Systems). The maximum speed of leaf motion is 2.5 cm/s. All plans were carried out with fixed jaw technique which keeps jaws in the same position during irradiation.
The dose prescribed to the PTV was 50.4 Gy in 28 daily fractions.
The planning goal was to achieve 50% or more of the PTV receiving the prescription dose, 95% of the PTV receiving 98% of the prescription dose, and then 0% of the PTV receiving 110% of the prescription dose. In addition, dose constraints for OARs were followed by the JCOG 1402 protocol. 11  Log files of the first fraction were analyzed for all patients.

2.D | Evaluation of mechanical control accuracy
Plan parameters of all IMRT plans were calculated by using inhouse software developed using the Eclipse scripting application programming interface (API), version 13.6. The mean aperture size and mean MLC gap width were calculated from the leaf positions for each control point.

2.E | Simulations of dosimetric uncertainties caused by patient setup errors
The simulation method was based on stochastic properties of rigid motions. 12 To investigate dosimetric uncertainties caused by patient setup errors, in-house software written by the Eclipse scripting API was used to perform the simulations. This study simulated residual setup errors after patient repositioning for CBCT-based IGRT. Planning CT was used to evaluate the effect of pure setup error without consideration of patient weight changes on dose delivery to CTV.
First, in-house software blurred the planned doses for rotational and translational setup errors in three dimensions. Previous study reported that patient setup errors after repositioning with a six degrees of freedom (6DOF) couch were 1.6±0.8 mm. 13 Therefore, planned doses were randomly blurred from 0 to 2 mm in translation and from 0°to 2°in rotation and simulated 28 times per patient. In UTENA ET AL.
| 109 this work, dose blurring means that the isodose cloud was randomly blurred. Therefore, this study did not consider inhomogeneity correction. Second, the maximum deviations in the CTV D 98% and D 2% between the planned doses and blurred doses were evaluated.

2.F | CBCT-based dose calculation
All patients underwent CBCT scans prior to every treatment. The Varian On-board Imager Spotlight protocol was used. 14  Moreover, there has been no report comparing the planned dose and a deformed accumulation dose of CBCT-based dose calculation for all fractions in the pelvic region. The purpose of this study was to evaluate the deformed dose with commercial software without performing any special image processing. Then, the deformed accumulated dose was evaluated only in patients who had small interfractional organ motion, CBCT images had no artifacts, and were able to contour OARs (bladder, rectum, and femoral heads) on CBCT images for all fractions. Fractional delivered dose was deformed to planning CT using deformed vector field (DVF) generated with RayStation ver.9.A (RaySearch Laboratories, Stockholm, Sweden). In this study, hybrid DIR algorithm that uses both intensity-based and structure-based DIR was used. 15 The online 3D/3D matching was used for an initial rigid registration. Deformation accuracy was evaluated using dice similarity coefficients (DSCs) of bladder, rectum, and femoral heads. 16 DSC measures the overlap volume between the ROI contoured by planning CT and the ROI deformed from CBCT to planning CT. DSC is widely used to evaluate deformation accuracy.
Finally, the deformed fractional delivered dose was accumulated.
Deformed accumulated doses were compared with the planned doses for D 98% and D 2% of CTV and D mean of OARs.
For dose comparisons of planned doses and CBCT-based delivered doses, Student's paired t test was used with R version 3.6.0 software (R Foundation, Vienna, Austria). Statistical significance was set at the 5% level, P < 0.05.

3.A | Evaluation of mechanical control accuracy
From log file analysis, the MLC positional errors and MU delivery errors were −0.01 ± 0.02 mm and 0.10 ± 0.10 MU, respectively.
The mechanical accuracy of TrueBeam was found to be well below the recommended tolerances by AAPM TG142. 17 From the analysis of plan parameters, the mean aperture size, mean differences in MLC gap width, and mean total MU were 34.4 ± 20.5 cm 2 , 16.9 ± 9.9 mm, and 1528 ± 117 MU, respectively. Table 1 shows the maximum differences between the blurred dose and planned dose in each patient. For all patients, the differences in the DVH parameters (D 98% and D 2% ) from the original plan were 0.5% or less. Residual setup errors ≤2 mm and 2°did not affect the target coverage.

3.C | Evaluation of fractional CBCT-based delivered dose
To investigate the effect of interfractional rigid or nonrigid geometric variations on the CTV dose coverage, the planned dose and CBCT-T A B L E 1 Comparison of DVH parameters (D 98% and D 2% ) between the blurred dose and planned dose. 3.D | Evaluation of accumulated CBCT-based delivered dose After the visual assessments, DIR was performed on 2 to 13 patients who were able to be contoured without artifacts in all fractions (patients 2 and 8).  between the deformed accumulation doses and the planned doses were also within 1%. However, the dose difference of rectum in patient 2 showed the large difference (4.6%). In addition, DVH curves of patients 2 and 8 are shown in Fig. 4. In patient 8, the DVH curve was consistent between the two doses, and the DVH parameters were also consistent within 2% including OARs. In contrast to patient 8, patient 2 showed the inconsistency of DVH curve between the two doses.
3.E | Effects of patient weight changes and internal organ deformations   In contrast to the evaluations of fractional doses, Fig. 3 shows that the interfractional deformations of internal organs contributed little to the dose differences in the accumulated dose. We found that the factors reducing target coverage could be divided into T A B L E 3 Comparison of DVH parameters between the deformed accumulation dose and the planned dose for patients 2 and 8. random and systematic components. The random components were interfractional organ deformation and patient setup errors.
As shown in Fig. 5(a−c), patient 13 showed weight gain during treatment. This variation in patient's weight caused dose decreases in CTV. In addition, dose increases due to the appearance of the gas pocket in rectum was detected as shown in Fig. 5(d−f). Previous study has also reported the increase in dose due to the gas pocket effect. 30  To ensure delivery of the prescribed dose for a patient, adaptive radiation therapy (ART) can be considered to prevent or minimize this dosimetric errors caused by patient body weight changes. A previous study discussed the importance of ART for head-and-neck cancer. 31 Although the ART technique was generally focused on headand-neck cancers, the study showed that the ART technique could be applied to the pelvic region. We think that evaluation of fractional delivered doses should be an essential quality assurance technique used during the whole treatment course.
In this study, the HU-to-relative electron density (RED) conver- They reported that geometric uncertainty of cervix caused by interfractional motion was larger than that of intrafractional motion.
Since the accumulated dose evaluation of CTV D 98% indicated that interfractional delivery errors were within 1.7%, the effect of intrafractional motion is considered to be even smaller.
Hence, DIR should be introduced for more accurate evaluation of fractional and accumulated dose differences in tumors and OARs. 36 Fig. 4(a)]. Patient 2 also had large dosimetric errors of CTV in Figs. 1 and 2. Figure 4 and Table 3 indicate that reproducibility of organ volume and gas pockets varies from patient to patient.
Although the deformed accumulation dose was evaluated only for two patients without artifact in all fractions, as shown in Table 2, deformation accuracy of hybrid DIR was smaller than previous study. 15 To improve accuracy of evaluation for accumulated delivered doses of OARs, generating synthesized CT image based on CBCT by a machine learning algorithm 33,34 will be introduced for reducing artifact effects in future work. Then, actual delivered doses of OARs will be analyzed with synthesized CT generated by CBCT and hybrid DIR technique in next step.
The same trend was found in dosimetric errors of CTV accumulated by rigid registration (Figs. 1 and 2) and dosimetric errors of CTV and OARs accumulated by DIR (Fig. 4). Then, this study shows the possibility of evaluating accumulated dosimetric errors of CTV Kershaw L et al reported that position errors of regional lymph nodes were inconsistent with bone matching. 25 Although the residual setup errors in bone matching were different for tumor and lymph node, these setup errors were sufficiently covered by PTV margin.
This work clarified that our institution could sufficiently deliver prescription dose to CTV. Although enough margin was added to CTV, interfractional variations of patient weight changes and gas pocket caused large dose differences (~5%) seen in Fig. 5.
Not only expanding PTV margin, adaptive radiation therapy will be required to eliminate these factors related to daily patient physiological changes. Jensen et al reported feasibility of adaptive strategy for cervical cancer therapy by daily CBCT-based monitoring by radiation therapists. 37 This study showed that monitoring fractional delivered doses helps radiation oncologists and medical physicists in decision making of replanning.
In conclusion, three categories of factors that contribute to decreased dose delivery accuracy were evaluated in this study.
We found that patient weight variations and internal organ deformations caused large target dose differences from the original plan doses. Monitoring of delivered doses should be added to a clinical workflow to periodically evaluate the delivered plan quality.

ACKNOWLEDG MENTS
We wish to acknowledge valuable discussion and excellent assistance with the member of Department of Radiation Oncology of Juntendo University Hospital.

CONFLI CT OF INTEREST
There is no conflict of interest related to this study. | 115

FUNDING INFORMATION
This study was not funded.