Impact of the MLC leaf‐tip model in a commercial TPS: Dose calculation limitations and IROC‐H phantom failures

Abstract Purpose Treatment planning system (TPS) dose calculation is sensitive to multileaf collimator (MLC) modeling, especially when treating with intensity‐modulated radiation therapy (IMRT) or VMAT. This study investigates the dosimetric impact of the MLC leaf‐tip model in a commercial TPS (RayStation v.6.1). The detectability of modeling errors was assessed through both measurements with an anthropomorphic head‐and‐neck phantom and patient‐specific IMRT QA using a 3D diode array. Methods and Materials An Agility MLC (Elekta Inc.) was commissioned in RayStation. Nine IMRT and VMAT plans were optimized to treat the head‐and‐neck phantom from the Imaging and Radiation Oncology Core Houston branch (IROC‐H). Dose distributions for each plan were re‐calculated on 27 beam models, varying leaf‐tip width (2.0, 4.5, and 6.5 mm) and leaf‐tip offset (−2.0 to +2.0 mm) values. Doses were compared to phantom TLD measurements. Patient‐specific IMRT QA was performed, and receiver‐operating characteristic (ROC) analysis was performed to determine the detectability of modeling errors. Results Dose calculations were very sensitive to leaf‐tip offset values. Offsets of ±1.0 mm resulted in dose differences up to 10% and 15% in the PTV and spinal cord TLDs respectively. Offsets of ±2.0 mm caused dose deviations up to 50% in the spinal cord TLD. Patient‐specific IMRT QA could not reliably detect these deviations, with an ROC area under the curve (AUC) value of 0.537 for a ±1.0 mm change in leaf‐tip offset, corresponding to >7% dose deviation. Leaf‐tip width had a modest dosimetric impact with <2% and 5.6% differences in the PTV and spinal cord TLDs respectively. Conclusions Small changes in the MLC leaf‐tip offset in this TPS model can cause large changes in the calculated dose for IMRT and VMAT plans that are difficult to identify through either dose curves or standard patient‐specific IMRT QA. These results may, in part, explain the reported high failure rate of IROC‐H phantom tests.


| INTRODUCTION
Installation of a new treatment planning system (TPS) requires rigorous commissioning and validation testing to ensure dose calculation accuracy. 1 This includes external validation of the dosimetry as recommended by various national and international groups [1][2][3][4] and as required for clinical trials. 5 Intensity-modulated radiation therapy (IMRT) credentialing is offered by the Imaging and Radiation Oncology Core Houston branch (IROC-H) in the form of anthropomorphic phantoms for several anatomical sites. 6 Recent results from IROC-H show that 17% of institutions using the service had considerable dose calculation errors in their TPS 7 and that patient-specific IMRT quality assurance (QA) at many of these institutions failed to predict these errors. 8 Despite significant evidence that patient-specific IMRT QA fails to detect TPS errors, [9][10][11][12] many institutions continue to rely on it for TPS commissioning and validation.
Although these failure patterns are established and appreciated in the community, it is not well-known which specific factors within the TPS may cause these failures, especially for newer planning systems. Small changes in the multileaf collimator (MLC) position are known to translate into large dose deviations in IMRT plans. [13][14][15][16] Previous studies have examined the potential impact of MLC miscalibration both in delivery 17 and as seen in log files. 18 Studies have also examined the dosimetric leaf gap (DLG) parameter used in the Eclipse TPS (Varian Medical Systems, Palo Alto, CA) to describe leaf offset positions, showing potentially large dosimetric impacts. [19][20][21] However, these studies do not explore whether these results translate across planning systems, nor do they assess the detectability of these deviations with QA devices or whether anthropomorphic phantom tests may aid in the detection of inaccurate MLC modeling.  22 were performed and were found to agree to within the recommended tolerances.
External validation was carried out through irradiation of the IROC-H head-and-neck phantom. 23 The phantom contains   eight total TLDs, labeled here as PTV1 center, PTV1 periphery,   PTV2, and Spinal cord, each with both superior and inferior TLDs. For simplicity and clarity, this study only includes the superior TLDs, as doses from the superior and inferior TLDs were found to agree within 0.3% in the PTVs and 1.5% in the spinal cord.
The phantom was scanned on a CT simulator following standard clinical workflow, and nine uniquely optimized treatment plans were created in the TPS following the planning guidelines given by IROC-H. Of these plans, five plans were step-and-shoot IMRT plans containing either seven or nine beams evenly distributed around the patient. Note that Elekta linear accelerators do not allow dynamic IMRT beam delivery except for VMAT plans. The four remaining plans were delivered using VMAT containing either two or three full arcs with variable collimator angles, dose rate, and gantry speed. The total monitor units (MU) for all plans ranged between 1469 and 2114 MU. Table 1 summarizes beam arrangement and MU information for each plan. All plans were optimized by prioritizing different optimization objectives, while still remaining within the IROC-H planning guidelines, in order to obtain unique optimization solutions. All plans were typical of how our institution would treat a head-andneck patient, both in terms of complexity and beam arrangements.
One of these plans, a 7-field step-and-shoot IMRT plan (Plan IMRT0 in Table 1), was delivered to the IROC-H phantom after standard patient-specific QA and physics checks were performed.  Five step-and-shoot IMRT plans and four full arc VMAT plans were individually optimized to ensure a variety of plans with unique solutions were investigated.   Each leaf-tip offset shows two boxplots, one for IMRT (5 plans) and one for VMAT (4 plans). The clinical model, with a leaf-tip offset of −0.5 mm, was used as a baseline. Thus, the dose difference at −0.5 mm leaf-tip offset is equal to zero.
In general, a more negative leaf-tip offset value, which corresponds to a smaller field size, underestimates the dose, while a more positive leaf-tip offset, which corresponds to a larger field size, overestimates the dose. The results are approximately symmetric around the clinical model, though the exact relationship depends on the specific beam parameters of each plan. This can be seen in Figure 4 for all four TLDs, with models that have a leaf-tip offset that is more negative than the clinical model underestimating the dose, and models with a more positive leaf-tip offset overestimating the dose.
Similar to the results seen in Figure 3,

| CONCLUSION S
The MLC modeling parameters in a commercial TPS were investigated in the context of IMRT treatment planning. Changes in these parameters had large effects on IMRT dose, but these effects were not readily apparent during the standard modeling, commissioning, and validation processes. This effect was much more pronounced for the leaf-tip offset than the leaf-tip width, with dose differences up to 20% for a 1 mm shift in the MLCs.
IMRT QA was unable to detect failing models unless the dose deviation was very large (>20%). Only external validation with an anthropomorphic phantom was able to reliably detect failing models. Care should be taken during the modeling process of the MLCs in a new planning system, and external audits recommended by national and international societies are an essential component of safe TPS commissioning.

CONFLI CT OF INTEREST
No conflicts of interest.