Introduction
The implementation of stereotactic radiosurgery (SRS), stereotactic radiotherapy (SRT), and stereotactic body radiotherapy (SBRT) has always been associated with high demands on dosimetry for the accurate and safe delivery of the corresponding treatments [
1‐
3].
The Monte Carlo (MC) method plays a key role exploiting the statistical nature of the photons’ and their secondary particles’ interactions. It is generally considered as golden standard for the fundamental investigation of particle interaction processes, relevant for both measurement and calculation of dose distributions. As an example, the MC method is effectively used to determine measurement based correction factors, which are crucial especially in small field dosimetry [
4‐
6].
One of the drawbacks of MC based solutions is the fact that MC methods are very computationally expensive. Depending on parameters like the intended statistical uncertainty, size of the problem (i.e. voxel size, calculation volume, etc.), it might be necessary to spend hours for the computation of MC based dose distributions, affecting the clinical transferability of MC methods. However, our group developed several strategies to overcome this limitation without making unacceptable compromises in terms of accuracy [
7‐
10]. With respect to stereotactic treatments, we recently developed a vendor independent dose calculation (IDC) framework for the calculation of dose distributions for the CyberKnife® M6 radiosurgery system (Accuray Inc., Sunnyvale, CA) equipped with the InCise™ multileaf collimator (MLC) [
11]. The IDC framework has been validated against measurements and showed only small differences in the order of 2% dose difference or 2 mm distance to agreement between calculated and measured dose distributions. As a consequence, the IDC framework serves not only as a highly accurate method for dose calculation, but is also useful for verification purposes. It may be possible to reduce the clinical workload for patient specific quality assurance (QA) by replacing cumbersome measurements with IDC.
More recently, a new MC based dose calculation method was released by Accuray as part of the TPS for stereotactic treatment planning purposes using the CyberKnife M6 equipped with the InCise MLC. While MC based dose calculation has been available for CyberKnife treatments using Cone and Iris collimators for several years [
12,
13] and is well validated [
14‐
16], no such validation exists for the newly introduced MC algorithm for MLC treatments. In the following, we refer to this implementation as Precision MC. The aim of this work is to benchmark this commercially available dose calculation algorithm by comparing Precision MC calculated dose distributions and resulting dose volume histogram (DVH) based parameters with the corresponding results using the IDC framework. For this purpose, the accuracy, efficiency, and usability of Precision MC are examined in academic and clinically motivated situations.
Discussion
In this work, a commercially available dose calculation algorithm (Precision MC) for CyberKnife M6 treatments was benchmarked against measurements as well as against an independent dose calculation framework (IDC). A prior benchmarking study of the Precision MC for MLC included single beam tests in which Precision MC and FSPB calculations were compared to film and chamber measurements in a heterogeneous slab phantom [
18]. That study showed good agreement (2%/1 mm 2D gamma passing rates of 91.2 ± 1.5%) between Precision MC and measurement in very low density lung substitute materials with localized anomalies due to a simplification in electron transport, which improved further after a modification was made to the electron transport algorithm (gamma passing rates of 96.6 ± 1.2%). That modification was introduced by the manufacturer before work for the current study started and is included in the Precision MC version used in our study. The work concentrated on MLC only, since MC-based methods are already existing for the other two collimator options (i.e. fixed size cones, IRIS) for the CyberKnife system. For this benchmark, different complexity levels were considered by looking at simple cases such as a homogeneous water tank, phantom cases reflecting lung or pelvis treatments, and clinical cases (all lung). By this means, it was possible to comprehensively investigate an entire spectrum of situations, which are not only of physics interest but also of clinical relevance.
Generally, for the cases considered, there is good agreement between Precision MC, IDC, and measurements. This was quantified by dose difference, distance to agreement, gamma evaluation, and DVH analyses (Tables
3 and
4).
The work also indicates the difficulty of accurately handling small fields in radiation therapy. Although IDC is based on the MC method – generally known as the most accurate dose calculation method – we observed dosimetric differences between IDC and measurements in the order of 4% for the smallest field. Since for the clinical cases, the treatment plans also include small fields, the observed differences for the clinical cases can partly be associated to this effect. Part of these differences can also be attributed to effects of partial source obscuration in the IDC MC model for the smallest field.
While for two prostate phantom treatment plans and seven clinical (lung) treatment plans the gamma evaluation between Precision MC and IDC results in passing rates > = 98.1%, we also showed the relevance of using MC instead of FSPB for heterogeneous anatomical situations. The addition of MC based dose calculation to MLC based CyberKnife treatments was thus important for treatments in regions such as lung, liver and mediastinal tumors.
Nevertheless, even though both algorithms – Precision MC and IDC – are based on the MC method, residual differences in the calculated dose distributions remain, which cannot be explained by statistical uncertainty only. Besides the different beam models used, the underlying MC transport methods differ (proprietary code and EGSnrc, respectively) and thus serve as an explanation for the observed differences. Moreover, as outlined in this work, both material conversion and electron track generation is handled differently in the two algorithms. Material conversion for photon interaction simulation on one hand, is simplified in Precision MC, assigning one of three materials (air, soft tissue or bone) to voxels instead of the 14 stoichiometric material compositions used in IDC. For electrons, on the other hand, interactions are pre-simulated in water. This leads to dose differences near inhomogeneous tissue (air/soft tissue/bone interfaces), which consequently are then observed in the corresponding phantom and clinical cases. As shown in Figs.
5b and
6b, Gamma values peak in regions of tissue interfaces (e.g. bronchi, pleura or the surface of the vertebral body) with IDC showing lower dose than Precision MC in air close to soft tissue and higher dose in soft tissue close to bone. These dose differences appear to be confined to small regions, suggesting the electron pre-simulation in water being the underlying cause. Even though peak differences in the dose of 2.5% were observed, these are generally acceptable differences for clinical routine work with gamma passing rates for 2% / 1 mm of 98.1% or greater [
1].
Calculation efficiency of Precision MC is optimized (whereas in IDC it is deliberately not) and compares favorably to similar described frameworks [
21]. This is to be expected, as IDC serves a “gold standard” purpose, not employing any efficiency improving approximations. While IDC takes 4–6 h to calculate dose to clinically acceptable mean statistical uncertainties of about 2% [
11], Precision MC does so within 2.2–8.1 min at native CT resolution (note that if the resolution is reduced to 256 × 256 x number of slices, these times reduce to 41–132 s).
Both accuracy and efficiency of the Precision MC dose calculation are within clinically accepted limits rendering the system practical for routine use.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.