Does intensity modulation increase target dose calculation errors of conventional algorithms for lung SBRT?

Abstract Purpose Conventional dose algorithms (Type A and Type B) for lung SBRT can display considerable target dose errors compared to Type‐C algorithms. Intensity‐modulated techniques (IMRT/VMAT) are increasingly being utilized for lung SBRT. Therefore, our study aimed to assess whether intensity modulation increased target dose calculation errors by conventional algorithms over conformal techniques. Methods Twenty lung SBRT patients were parallely planned with both IMRT and dynamic conformal arc (DCA) techniques using a Type‐A algorithm, and another 20 patients were parallely planned with IMRT, VMAT, and DCA using a Type‐B algorithm. All 100 plans were recalculated with Type‐C algorithms using identical beam and monitor unit settings, with the Type‐A/Type‐B algorithm dose errors defined using Type‐C recalculation as the ground truth. Target dose errors for PTV and GTV were calculated for a variety of dosimetric end points. Using Wilcoxon signed‐rank tests (p < 0.05 for statistical significance), target dose errors were compared between corresponding IMRT/VMAT and DCA plans for the two conventional algorithms. The levels of intensity modulation were also evaluated using the ratios of MUs in the IMRT/VMAT plans to those in the corresponding DCA plans. Linear regression was used to study the correlation between intensity modulation and relative dose error magnitudes. Results Overall, larger errors were found for the Type‐A algorithm than for the Type‐B algorithm. However, the IMRT/VMAT plans were not found to have statistically larger dose errors from their corresponding DCA plans. Linear regression did not identify a significant correlation between the intensity modulation level and the relative dose error. Conclusion Intensity modulation did not appear to increase target dose calculation errors for lung SBRT plans calculated with conventional algorithms.


| INTRODUCTION
Stereotactic body radiotherapy (SBRT) is an increasingly common treatment for patients with inoperable early stage non-small cell lung cancer (NSCLC) and other lung tumors, with well-demonstrated efficacy and minimal side effects. [1][2][3] Accurate dose calculation during treatment planning is especially important in SBRT, owing to highfractional doses delivered over a small number of fractions. However, two factors unique to lung SBRT increase the difficulty of achieving this necessary accuracy. First, tissue heterogeneity between the high-density tumor and the surrounding low-density lung tissue complicates dose calculations owing to the loss of charged particle equilibrium. 4 Second, the small field sizes associated with the small target volume further exacerbate this problem. [5][6][7] The obstacle of dose calculation has been addressed by the development of more accurate dose calculation algorithms in commercial treatment planning systems, from earlier homogeneous dose calculations to newer dose algorithms incorporating heterogeneity correction. 4 These latter algorithms are usually categorized from Type A to Type C, with increasing dose calculation accuracies: 8,9 (1) Type A: algorithms with a one-dimensional equivalent path length correction such as pencil beam (PB) convolution and ray tracing; 5,7,10 (2) Type B: algorithms applying two-dimensional corrections such as collapsed cone convolution (CCC) 11 and analytical anisotropic algorithm (AAA); 12 and (3) Type C: advanced algorithms such as fast Monte Carlo algorithms 4,13 and Boltzmann Solver-based algorithms such as acuros external beam (AXB). 9 Dose calculation has been widely compared among the above algorithms for lung SBRT, with interalgorithm differences mostly observed for the target dose. 5,7,9,10,14 It is known that as compared with Type-C algorithms, Type-A and Type-B algorithms tend to overestimate the target dose for lung SBRT, and the magnitude of the dose error varies widely from case to case, up to over 30% for Type A and over 15% for Type B. 7,9,15 These results include both forward plans using conventional conformal techniques as well as inverse plans using modern intensity-modulated techniques. Early lung SBRT primarily employed conformal techniques such as the 3D conformal beam technique and the dynamic conformal arc technique. 1,16 However, in recent years, intensity-modulated techniques such as intensity-modulated radiation therapy (IMRT) and volumetric-modulated arc therapy (VMAT) have gained increasing popularity owing to superior organ-at-risk sparing, dose conformity, and fast delivery in the case of VMAT. 9,17 Also contributing to these paradigm shifts where recent studies displaying that the dominant effect of the "interplay" between tumor/tissue motion and multileaf collimator (MLC) modulation involved the blurring of dose distributions, which appeared to be small. 18,19 However, other concerns remain regarding the suitability of intensity-modulated techniques for lung SBRT treatments, associated with smaller field apertures, and specifically whether they would lead to significantly higher dose errors than conformal techniques when calculated with conventional dose algorithms. This concern was based on previous findings that field size impacts dose errors (wherein smaller target sizes were associated with higher dose errors than larger target sizes). 5,7,14,20 However, this intuitive concern has not been substantiated; the dose error influence of intensity modulation, which effectively generates smaller field apertures, remains to be elucidated.
In this study, we designed a back-to-back comparison to study the impact of intensity modulation on dose errors of conventional Type-A and Type-B algorithms. The target dose errors of a Type-A algorithm were statistically compared between IMRT and conformal techniques, and those of a Type-B algorithm were compared between VMAT, IMRT, and conformal techniques.

2.A | Patient simulation and contouring
Under the approval of the University of Nebraska Medical Center Institutional Review Board, 40 patients with early stage non-small cell lung cancer (NSCLC) treated with lung SBRT at our institution between June 2012 and August 2016 were randomly selected for this retrospective study. The simulation and contouring process was as previously described. 15 Briefly, patients were simulated with a free-breathing 3D CT followed by a 4D CT. The gross tumor volume (GTV) was delineated using only the 3D CT, while the internal target volume (ITV) was delineated using both the 4D and 3D CTs, and the planning target volume (PTV) was generated by adding an isotropic expansion of 5 mm to the ITV. For the Type-A study, a dynamic conformal arc (DCA) plan as previously described and an IMRT plan were created for each patient using a pencil beam algorithm with equivalent path length heterogeneity correction in iPlan v4.5 (Brainlab AG, Feldkirchen, Germany). 15 Both DCA and IMRT plans used a 360°geometry, with a full arc for DCA and seven equal angularly spaced beams for IMRT.

2.B | Patient grouping and treatment planning
Dose errors were compared between the two plan modalities for the Type-A algorithm, using recalculated dose (with identical beam and monitor unit settings) by the Voxel Monte Carlo (VMC) algorithm in iPlan v4.5 as the ground truth. For VMC calculation, the full MLC ZHENG ET AL. | 155 geometry simulation "Accuracy Optimized Model," with a spatial resolution of 2 mm and variance of 1%, was used.
For the Type-B study, a DCA plan, a VMAT plan as previously described, 9 and an IMRT plan utilizing the same planning objectives were created for each patient using the AAA algorithm in Eclipse v13.5 (Varian Medical Systems, Palo Alto, CA, USA). Most patients had full-arc DCA and VMAT plans, together with seven-beam IMRT plans equally spaced over 360°. Two patients with very peripheral lesions had partial-arc DCA and VMAT plans, together with sevenbeam IMRT plans equally spaced over the same partial arc. Dose errors were compared between the three plan modalities for the Type-B algorithm, using recalculated dose (with identical beam and monitor unit settings) by the AXB algorithm in Eclipse as the ground truth.
A dose grid of 2 mm was used for all of the above calculations.
Dose to medium was used for both VMC and AXB calculations.   Table 1. A median modulation ratio of 1.4, 1.4, and 1.5 was calculated for Type-A IMRT plans, Type-B VMAT plans, and Type-B IMRT plans, respectively.

3.B | Type-A dose errors
As listed in Table 2, the Type-A calculation led to a target dose overestimation for both the PTV and GTV for all dose end points on all plans.
The dose error magnitude varied widely from case to case, from a few percentage to over 30%. In general, the dose errors were larger for target "cold spot" metrics such as Dmin and D95% than for target "hot spot" parameters such as Dmax and D5%. They were also larger for the PTV than for the GTV. Comparing between IMRT and DCA plans, most dose end points displayed no statistically different dose errors.
For a couple end points with p<0.05, the absolute dose differences (where IMRT dose errors were numerically smaller than those for DCA) were very small and clinically insignificant.  Table 3.

3.C | Type-B dose errors
Comparing between IMRT and DCA or between VMAT and DCA, the "cold spot" parameters such as Dmin and D95% had, on

3.D | Intensity modulation vs relative dose errors
As listed in Table 1, the levels of intensity modulation were moderate for the 60 intensity-modulated plans investigated in the study, with modulation ratios between 1.2 and 2.2. The relative dose errors did not appear to correlate with the levels of plan intensity modulation. For example, a scatter plot is depicted in Fig. 1 to illustrate the relationships between the relative dose error of PTV D95% and the modulation ratio of the plan for the 60 intensity-modulated plans.
As indicated by the low R 2 values of the linear regression, a clear dependence between the relative dose errors and the plan modulation ratios was not observed. Our study also did not find any linear correlation between the plan modulation ratio and the relative dose error of the modulated plan over the conformal plan. There is likely a complex relationship between the specific beamlets in a modulated plan in terms of location and size and their dose error contributions due to heterogeneity. It is therefore speculated that a compounded and nonlinear relationship existed between the total relative dose error and the overall modulation ratio.

| DISCUSSION
The dose errors observed in this study for Type-A and Type-B algorithms are in agreement with numerous previous studies. 7,9,14,15 Both algorithms overestimated the target peripheral or "cold spot" indices such as D95% and Dmin, and the magnitude of the overestimation varied widely from case to case. The Type-A algorithm errors were in general larger up to over 30% for PTV D95%, making the algorithm unsuitable for lung SBRT planning. The Type-B algorithm errors were smaller for most cases, but could be up to 11% for some cases, which may necessitate the use of more accurate Type-C algorithms going forward.
In our study, the VMAT technique was only investigated for the Type-B algorithm (not for the Type-A algorithm) because the treatment planning system iPlan utilized for the Type-A investigation does not support VMAT planning. Another limitation of our study was that only one Type-A and one Type-B algorithm were investigated; therefore, the results and trends found may or may not be generalizable to other algorithms. Herein, the Type-B algorithm utilized was AAA, which may be less accurate than another Type-B algorithm, CCC. [24][25][26] However, it should also be noted that AAA is still quite prevalent in the current clinical treatment planning, which makes this study quite relatable to contemporary practice. Finally, our study used DCA as the conformal technique for comparison. While DCA is a popular conformal technique used for lung SBRT, 9,10,[14][15][16] other techniques are also used such as multiple noncoplanar conformal beams.

| CONCLUSION S
The impact of intensity modulation on conventional algorithm dose errors for lung SBRT was investigated herein. Intensity modulation does not appear to increase the target dose errors for lung SBRT plans calculated with conventional algorithms.

CONFLI CTS OF INTEREST
The authors have no relevant conflicts of interest to disclose.