Abstract
Background/Aim: This study evaluated the impact of knowledge-based plan (KBP) model improvement on plan complexity and delivery accuracy in volumetric modulated arc therapy (VMAT) for prostate cancer at multiple institutions. Materials and Methods: Five institutions created the first KBP model before April 2017 and subsequently devised a new model (second model) based on feedback from the first KBP and the efforts of planners after April 2019. The dose–volume histogram (DVH) parameters were validated for two prostate cancer cases between the first and second KBPs. Plan complexity metrics, of the modulation complexity score for VMAT (MCSv), closed leaf score (CLS), small aperture score (SAS), and leaf travel (LT), were compared. The delivery accuracy metrics of γ pass rate and point dose discrepancy (plan vs. measurement) at isocenter were also compared. Results: There were no significant differences in DVH parameters between the KBPs. Conversely, V50% of the rectum and bladder was reduced in 6/10 and 8/10 patients, respectively, and these variations were also converged from the first KBP to the second KBP. The mean±1SDs of MCSv, CLS, SAS20mm, and LT (first KBP vs. second KBP) were 0.27±0.033 vs. 0.26±0.044, 0.062±0.032 vs. 0.14±0.091, 0.59±0.048 vs. 0.70±0.14, and 411.91±32.08 mm vs. 548.33±127.50 mm, respectively. The delivery accuracy did not differ, whereas MCSv was moderately correlated with the point dose discrepancy. Conclusion: Multi-leaf collimator motion could be more complex with KBP model improvement, which had the potential to deteriorate the delivery accuracy.
Intensity modulated radiotherapy (IMRT) and volumetric modulated arc therapy (VMAT) using inverse planning techniques improve both target conformity and normal tissue sparing. However, the plan quality depends on planner and institutional experience and skill because of the complexity of IMRT/VMAT goals (1-3), causing large variation of the plan quality among planners or institutions. The commercial knowledge-based plan (KBP) module RapidPlan® (Varian Medical Systems, Palo Alto, CA, USA) has been released for clinical use with the Eclipse (Varian) treatment planning system. Many studies have reported that the KBP can generate better or at least comparable dose distribution to the clinical plan (CP) at some anatomical sites (4-15). KBPs are expected to reduce the variations in plan quality, thereby improving planning consistency (4).
Our previous multi-institutional study also reported the improved KBP quality and reduced variation for the second-generated KBP based on the KBP feedback and the planners’ efforts at each institution (16). However, the effects of KBP improvement on plan complexity and deliverability were unclear. Some studies found that the plan complexity was more complicated with a KBP than with a CP, increasing the monitor unit (MU) (10) and decreasing the delivery accuracy (17). Moreover, the plan complexity and delivery accuracy in the case of CP has also varied across different institutions (18, 19). Therefore, in the case of KBPs, variation can occur across different institutions because the KBP model reflects the characteristic of each institutional plan such as policies and dose constraints. The purpose of this study was to clarify the impact of KBP improvement by model updates on plan complexity and delivery accuracy at multiple institutions in two periods.
Materials and Methods
Institutions and plan design. Five institutions (A-E) with different VMAT planning policies for patients with T1-T2c prostate cancer were included in this study. The definitions of gross tumour volume, margins defining the clinical target volume (CTV), and planning target volume (PTV) in each direction and the dose constraints were described in a previous study (20). At each institution, the KBP model was created before April 2017, and the second KBP model was created after April 2019. For the KBP model, the numbers of registered cases at institutions A, B, C, D, and E were 123, 53, 20, 60, and 100, respectively, whereas in the second KBP model, 50, 50, 50, 60, and 34 cases were enrolled, respectively. The dose constraints at each institution are presented in Table I (16, 20). The Institutional Ethics Committee approved this study (Kindai University review board No. 31-273).
KBP model validation. The dose distributions with the first and second KBP models were validated for two sets of CT data and structures of patients (Cases 1 and 2) at institution B. The data were anonymised and delivered to other institutes (16, 20). The thickness of the CT images was 2.5 mm, the field of view was 50 cm, and the target and organs at risk were contoured by an expert radiation oncologist according to the protocol of institution B (16, 20). At each institution, the planners who participated in this study had experience with inverse planning for IMRT or VMAT using the Eclipse 13.0 or 15.5 treatment planning system (TPS). Single optimisation was performed with each model for the validation. Photon Optimizer ver. 13.0 or 15.5 (Varian) was used for optimisation, and the calculation algorithm was the Anisotropic Analytical Algorithm (Varian) with grid size of 2.5 mm (16, 20).
The following dose–volume parameters were compared between the first and second KBP models: (a) dose received by 95.0% of the PTV (D95%); (b) dose received by 2.0% of the PTV (D2%); (c) homogeneity index [HI; defined as HI=(D2%-D98%)/D50%; where D2%, D50%, and D98% are the minimum doses to 2.0%, 50.0%, and 98.0% of the PTV, respectively] (2); (d) dose–volume parameters to the rectum as V90%, V80%, and V50%; and (e) dose–volume parameters to the bladder as V90%, V80%, and V50% (16, 20).
Plan complexity and delivery accuracy of each KBP. Plan complexity was verified for the two cases of VMAT with the first and second KBP models using the following metrics (3):
Mean field area (MFA): Mean field area weighted the MU at each control point (21, 22).
Closed leaf score (CLS): Proportion of closed multi-leaf collimator (MLC) leaf pairs in-field (21, 22).
Small aperture score (SAS): Proportion of open MLC leaf pairs separated by less than the given thresholds (2, 5, 10, and 20 mm in this study) in-field (21, 22).
Leaf travel (LT): Mean distance of leaf movement in-field (23).
Modulation complexity score for VMAT (MCSv): Sum of all control points of the product of aperture area variability, leaf sequence variability, and normalised MU. MCSv assesses the variability between multi-leaf collimator positions and the aperture opening, and the range of values is 0-1 (23).
MU: Sum of all MUs for a plan.
High CLS, SAS, LT, and MU and low MFA and MCSv indicated more complex MLC motion. All plan complexity metrics were calculated with MATLAB software (MathWorks, Natick, MA, USA) using Digital Imaging and Communications in Medicine files.
To evaluate the delivery accuracy of each KBP, we performed a multi-institutional patient-specific quality assurance (QA) analysis for two VMAT cases using the first and second KBP models and the same experimental setup. The treatment equipment, MLC type, and measurements devices of the point dose and dose distribution for patient-specific QA in each institution are listed in Table II. The point dose at isocenter was measured with each ionisation chamber, as presented in Table II. The collecting volume of the ionisation chamber was contoured on the institution’s water equivalent CT value. The planned dose was defined as the average dose in this volume. The measured dose was corrected by the daily output factor to reduce the effects of daily linac output variations and differences between the phantom and liquid water. Dose distributions were also measured using the measuring system (Table II) and methodology as per the usual practice at each institution. These values were compared with the dose distributions calculated by the TPS. The differences between the measured and calculated dose distributions were evaluated using the passing rate of γ index with three tolerances in terms of dose differences and distance to agreement (2%/1 mm, 2%/2 mm, and 3%/3 mm) with thresholds of 30% and 60%. In institutions A and E, the relative dose distributions normalised to the dose at isocenter were evaluated, whereas at the other institutes, the absolute dose distributions were evaluated. Then, the relationships among the dose–volume parameters, plan complexity score, and delivery accuracy were also evaluated.
Treatment equipment, multi-leaf collimator (MLC) type, devices for point dose and dose distribution verification, and acceptable criteria for patient-specific quality assurance (QA) at each institution.
Statistical analysis. Statistical analyses were performed using R ver. 3.4.2 (R Foundation for Statistical Computing, Vienna, Austria). Wilcoxon’s signed rank test was used to compare the continuous variables and trends, and p<0.05 was considered statistically significant. Pearson’s correlation coefficient was considered weak, moderate, and strong at <0.4, 0.4-0.7, and >0.7, respectively.
Results
Validation of the KBP. There were no significant differences between the first and second KBPs across all dose–volume parameters in Cases 1 and 2. On the other hand, V50% for the rectum was reduced by more than 5% between the first and second KBPs, whereas V90% and V80% for the rectum were largely unchanged. For the bladder, the dose–volume parameters displayed small differences between the first and second KBPs (16). Figure 1 presents the differences for V50% of the rectum and bladder between the first and second KBPs at each institution. The numbers of patients with improvements between the first and second KBPs were 8/10 for V50% of the rectum and 6/10 for V50% of the bladder. The variation of plan quality was also decreased from the first KBP to the second KBP, as presented in Figure 1.
Differences between the first and second knowledge-based plans (KBPs) for V50% of the rectum and bladder in Cases 1 and 2 at each institution.
Plan complexity and delivery accuracy of each KBP. Table III presents the plan complexity metrics for the first and second KBPs for Cases 1 and 2. LT was significantly greater for the second KBP than for the first KBP. CLS and SAS were also higher for the second KBP than for the first KBP, although the differences were not significant. SD was larger for the second KBP than for the first KBP for all metrics excluding MFA. Figure 2 presents the difference between the first and second KBPs for each plan complexity metric. The numbers of plans with more complex MLC motion for the second KBP than for the first KBP were 6/10, 7/10, 7/10, and 9/10 for MCSv, MU, SAS20mm, and LT, respectively. Contrary to the variation of dose–volume parameters, the change of plan complexity metrics, especially SAS20mm and LT, were larger for the second KBP than for the first KBP (Figure 2).
Plan complexity metrics (mean±SD) for the first and second knowledge-based plans (KBPs) for Cases 1 and 2.
Differences between the first and second knowledge-based plans (KBPs) in modulation complexity score for VMAT (MCSv), monitor unit (MU), small aperture score with threshold of 20 mm (SAS20mm), and leaf travel (LT) in Cases 1 and 2 at each institution.
The relationships between the differences of MCSv, MU, SAS20mm, and LT and those of V50% for the rectum and bladder between the first and second KBPs in Cases 1 and 2 are presented in Figure 3. Strong correlations were observed between the difference of MCSv and differences of V50% in the rectum and bladder, whereas moderate correlations were observed for the remaining metrics, indicating that plan quality improved with increasing plan complexity.
Relationships between the differences in (A, B) leaf travel (LT), (C, D) small aperture score with threshold of 20 mm (SAS20mm), (E, F) modulation complexity score for VMAT (MCSv), and (G, H) monitor unit (MU) and that of V50% to the rectum and bladder between the first and second models in Cases 1 and 2.
The γ pass rates and point dose discrepancies at each institution are presented in Table IV. The numbers of patients with γ pass rates exceeding 90% for the criteria of 2%/2 mm and 3%/3 mm with a threshold of 30% were 16/20 and 18/20, respectively. All point dose discrepancies passed each institutional criterion (Table II). The γ pass rates depended on the characteristics of each institution, which also means that plan complexity was not related to the γ pass rates. Conversely, the point dose discrepancy was moderately correlated with MCSv, as presented in Figure 4A, although the correlations between the other metrics (e.g., MU in Figure 4B) and the point dose discrepancy were weak.
The γ pass rates and point dose discrepancies at each institution.
Relationships of the point dose discrepancy with (A) modulation complexity score for VMAT (MCSv) and (B) monitor unit (MU).
Discussion
In this study, we investigated plan complexity and delivery accuracy for updated KBPs at several institutions in two periods. Our previous study revealed that the dose–volume parameters and dose distribution of KBP were improved by the model update (16). Conversely, LT increased significantly between the first and second KBPs, and the ratio of closed leaf or small MLC aperture size and the modulation complexity were also increased, thereby increasing MU. This explained why the improvements of the dose–volume parameters of KBP following the model update were moderately or strongly correlated with increases in LT, the rate of small MLC aperture size, modulation complexity, and MU, as presented in Figure 3. Thus, the improvement of the KBP quality resulted in an increase in plan complexity. The KBP quality at institutions A, B, and E was especially improved (Figure 1). Meanwhile, SAS20mm and LT were greatly increased (Figure 2), which increased the variation of plan complexity, whereas the variation of KBP quality was reduced.
The γ pass rates for 2%/2 mm exceeded 95% at institutions A, C, and E and 90% at institution B with a threshold of 30%. Most institutions passed the criterion of γ pass rate, as presented in Table II. At institution D, the γ pass rates of Cases 1 and 2 for the criterion of 3%/2 mm with a threshold of 30% were 73.9% and 73.5%, respectively for the first KBP and 88.8% and 92.8% for the second KBP, respectively. All point dose discrepancies passed each institutional criterion, as presented in Table II. However, the dose discrepancy increased with increasing MLC modulation complexity, as presented in Figure 4A. When patient-specific QA fails, there is more complex MLC motion than normal in clinical practice when using the KBP. Planners should consider replanning and attempt to achieve less plan complexity (24) [e.g., mitigate the priority of the objectives or using the aperture shape control tool (25-27)].
The KBP is useful for improving the dose–volume parameters and dose distribution and reducing their variation (4); conversely, plan complexity and its variation can increase, which can potentially deteriorate the delivery accuracy. Therefore, the plan complexity and delivery accuracy of the KBP must be validated carefully before clinical use as a part of the KBP validation. Appropriate cut-offs should be selected to understand the characteristics of plan complexity at each institutional KBP or clinical plan and experience such as the average complexity of similar plans, treatment site, and linac to increase the safety of the KBP procedure (28-30). Furthermore, in the case of treatment area with complex geometry such as the head and neck, the patient specific QA with a heterogeneous anthropomorphic phantom might improve the prediction of unacceptable plan delivery (31, 32). The use of a single model by several institutions [e.g., multi-institutional model (33-35)] might overcome the large intra-institutional variation of plan complexity caused by the improvement in plan quality.
Conclusion
MLC motion can become more complex with improvements of the dose–volume parameters of KBP following the model update, which can potentially decrease the delivery accuracy. A large variation of plan complexity could occur, whereas the dose–volume parameters of KBP were standardised at multiple institutions. Planners must validate the dose–volume parameters and dose distribution as well as plan complexity and delivery accuracy when using KBPs.
Acknowledgements
This work was supported by the Japan Society for the Promotion of Science KAKENHI (grant number: 21K07576). The Authors thank Joe Barber Jr., PhD, from Edanz (https://jp.edanz.com/ac) for editing a draft of this manuscript.
Footnotes
Authors’ Contributions
Concept and design: MT, HM. Treatment planning: MT, YU, JF, TK, YS, YM, KN. Measurements: MT, YU, JF, TK, YS, YM. Data analysis: MT, YU, KK. Manuscript preparation: MT, MH, HM. All Authors read and approved the final manuscript.
Conflicts of Interest
The Authors declare that they have no competing interests in relation to this study.
- Received August 3, 2022.
- Revision received August 28, 2022.
- Accepted August 29, 2022.
- Copyright © 2022 International Institute of Anticancer Research (Dr. George J. Delinasios), All rights reserved.
