A comprehensive dosimetric study on switching from a Type-B to a Type-C dose algorithm for modern lung SBRT

Lung cancer is the second most common type of cancer for both men and women, and the leading cause of cancer death, making up about 1 in every 4 cancer-related deaths [1]. Non-small cell lung cancer (NSCLC) makes up about 80–85% of all cases. Treatment for NSCLC varies depending on the cancer stage and other patient factors, but some common treatments include surgery, chemotherapy, immunotherapy, and radiation therapy. Stereotactic body radiotherapy (SBRT), also known as stereotactic ablative radiotherapy (SABR), is a type of external beam radiotherapy that is often used to treat early stage lung cancers as an increasingly popular alternative option to surgery [2‐4]. It is also often used to treat small-sized, oligo-metastasis in the lung [5].

Lung SBRT conventionally used multiple coplanar or non-coplanar conformal beams [6]. In recent years, volumetric-modulated arc therapy (VMAT) has gained greater popularity as the treatment technique of choice for lung SBRT because of the quick delivery time, superior critical tissue sparing and dose conformity, and robustness against the interplay effect due to respiratory motion [7‐12]. Several studies showed a substantial reduction in the treatment time for SBRT lung cases using VMAT when compared to intensity-modulated radiation therapy (IMRT) and conformal beam treatment, from dozens of to just a few minutes, especially when combined with the high dose rates provided by flattening-filter-free modes of modern linear accelerators (LINACs) [7, 9, 11]. While offering comparable steep dose gradients and critical tissue sparing to IMRT, VMAT was also found to be less susceptible to the interplay effect that prevented the wide application of IMRT in thoracic radiotherapy, especially when conventional dose rates are used [8, 10, 12]. However, accurate dose calculation in the presence of heterogeneity remains a challenge that arises when using SBRT [13‐18]. The heterogeneous tissue interfaces between low density lung tissue and high density tumor pose challenges for accurate dose modeling. Starting from homogeneous dose calculation, commercial treatment planning systems progressed in generations of heterogeneity corrected calculation algorithms, often classified as Type-A to Type-C, to better address this challenge [19‐21]. Conventional homogeneous dose calculations and Type-A or pencil beam algorithms with equivalent path length corrections can often lead to target peripheral dose over-estimation as high as 50% [22‐25]. As the current mainstay of commercial treatment planning algorithms, Type-B or convolution/superposition dose algorithms are commonly used due to their improved accuracy compared with older algorithms [17, 26]. On the other hand, studies have shown that while these algorithms resulted in <3% errors on the peripheral target dose in many patient cases, errors could be as high as over 10% compared with Monte Carlo calculation in other cases [27‐30]. Therefore, Type-C or fast Monte Carlo algorithms have been implemented in commercial treatment planning systems in the recent decade, which provide improved dose calculation accuracy over Type-B algorithms but faster dose modeling than full-fledged Monte Carlo simulation [29, 31‐33]. Acuros^® XB dose calculation algorithm, implemented for treatment planning in Eclipse (Varian Medical Systems, Palo Alto, CA), is based on a linear Boltzmann equation solver to solve radiation transport equation, rather than simulation of particle transport implemented in Monte Carlo. It is considered a Type-C algorithm due to the comparable level of dose modeling accuracy [30, 34‐36].

Many studies have investigated the dosimetric differences between Type-A or Type-B and Type-C dose algorithms, and have shown that Type-C algorithms more accurately calculate the dose distribution for SBRT plans of lung patients [15, 22‐24, 27, 32, 35]. Due to the large dose errors associated with Type-A algorithms, they have been gradually phased out in thoracic treatment planning. Modern lung SBRT protocols from National Research Group (NRG)/Radiation Therapy Oncology Group (RTOG) such as RTOG0813 and RTOG0915 require Type-B dose calculation [37, 38]. Despite the demonstrated dose calculation accuracy improvement and the increasing availability, currently the clinical utilization of Type-C algorithms is still limited [5]. In addition to technical challenges such as lengthier computation time and sub-optimal or lack of integration into inverse optimization engines compared with Type-B algorithms, there are also some clinical challenges. Firstly, inhomogeneous dose errors were revealed for Type-B compared with Type-C algorithms for different patients. While large errors can result for some patients, small errors within 3% were found for other patients [27‐30], suggesting that elaborate dose calculation using Type-C algorithms may not be necessary for every lung SBRT patient to offset the lengthier computation time. Secondly, current clinical guidelines on dose prescription and constraints have been established for Type-B algorithms and may not be directly applicable for Type-C algorithms.

Several authors have previously examined the dosimetric compliance of Type-C calculated lung SBRT plans to the current protocols [27, 29, 39, 40]. However, discordant recommendations were reported by these studies. Li et al. applied Monaco (Elekta/CMS, Crawley, UK), a Type-C algorithm, on 15 patients planned using XiO (Elekta/CMS, Crawley, UK), a Type-B algorithm, to examine RTOG 0813 compliance [29]. They recommended adjusting the dosimetric criteria (such as R_50%) because their values were increased with the Type-C algorithm and the corresponding compliance was worse. In contrast, Rana et al. published a study on 14 patients evaluating the effect of switching from a Type-B algorithm (Analytical Anisotropic Algorithm, or AAA) to a Type-C algorithm (AXB) in Eclipse for lung SBRT plans using RTOG0813 dosimetric criteria [27]. They found that the AXB re-calculation of the original AAA plans led to lower average values for most dose constraint parameters and reduced instances of protocol minor deviations on these parameters. Two studies by Pokhrel et al. used another Type-C algorithm, Voxel Monte Carlo implemented in iPlan (BrainLab AG, Feldkirchen, Germany), to evaluate the plan compliance with dosimetric criteria of RTOG0813 and RTOG0915 for central and peripheral lung SBRT patients, respectively [39, 40]. Based on the average values obtained on 20 patients, Pokhrel et al. recommended the necessary adjustments to the dosimetric criteria in order to make their Type-C plans fully compliant with the protocols.

The aim of this study, therefore, was to systematically investigate the necessity and dosimetric impacts of switching from Type-B to Type-C dose algorithms for lung SBRT planning on a large patient cohort. On fifty-two patients, the target dose parameters as well as RTOG dosimetric criteria were compared among the following plans: original Type-B plans, re-calculated Type-C plans with identical beam settings, re-normalized Type-C plans (to ensure target coverage), and re-optimized Type-C plans (to restore the original compliance to the dosimetric criteria achieved by the Type-B plans).

As will be detailed in the following sections, the dosimetric impacts of switching from Type-B to Type-C calculation were comprehensively evaluated using various dose-volume endpoints and protocol dosimetric constraints on 52 lung SBRT patient plans. We compared the original Type-B plans with the re-calculated, re-normalized, and re-optimized Type-C plans.

Our study investigated target dose, RTOG compliance, and other plan quality indexes on 52 lung SBRT patients by comparing between the original Type-B plan, the re-calculated Type-C plan, the re-normalized Type-C plan, and the re-optimized Type-C plan. Although previous studies in the literature have examined dosimetric differences arising from re-calculating Type-B treatment plans with the Type-C algorithm, discordant recommendations have been made regarding the applicability of current RTOG dosimetric criteria on Type-C calculated lung SBRT plans. Specifically, the study by Li et al. using Monaco [29] and the studies by Pokhrel et al. using Voxel Monte Carlo [39, 40] suggested that the RTOG dosimetric criteria, such as R_100%, R_50%, and D_2cm, might need to be up-adjusted to be more loose. In contrast, the study by Rana et al. using AXB showed comparable or even better compliance for the above criteria [27].

Our study shed some light on the conflicting results above. Similar to Rana et al.’s study [27], the re-calculated Type-C plans also showed comparable or even better compliance to RTOG dosimetric criteria, such as R_100%, R_50%, and D_2cm. However, what their study failed to consider was that Type-C re-calculation using the same MLC patterns and MUs would also lead to insufficient target dose coverage and hence sub-optimal plan quality for many patients. In fact, about 13.5% of our patient cohort showed a PTV D_95% reduction over 5%, and about 34.6% showed a PTV V_100% reduction over 5%. When the target coverage loss was made up by re-normalizing the MUs, not surprisingly, compliance to the above RTOG dosimetric criteria worsened for some patients (about 21.2% of our cohort) compared with the original Type-B plans. However, re-optimization using the Type-C algorithm was able to restore the original RTOG compliance for all but one of these patients. Current RTOG dosimetric criteria hence appear to remain appropriate for lung SBRT plans calculated with Type-C algorithms, provided that such algorithms are employed for plan optimization.

Pokhrel et al. recommended up-adjustments on the values for RTOG dosimetric criteria, such as R_100%, R_50%, and D_2cm, by about 10% in order for their Type-C plans to comply with the criteria [39, 40]. However, this was based on the premise that all plans need to fully comply with the criteria. Since in their studies the patients were planned with the Type-C algorithm, and no comparison was made with Type-B plans, it was not considered that for some of these 20 patients, RTOG criteria deviations could have resulted even if planned using Type-B algorithms. In fact, it is not uncommon in current clinical practice for Type-B lung SBRT plans to have minor or even major deviations from these RTOG protocols due to specific patient and tumor anatomy. For example, in the study by Rana et al., among the 14 clinical Type-B plans, 6 plans recorded minor deviations for R_100%, 9 plans recorded minor deviations for R_50%, 4 plans recorded minor deviations for D_2cm, and 1 plan recorded minor deviation for V₂₀. Similarly, in Li et al.’s series of 15 patients, Type-B plans also resulted in deviations on these RTOG criteria for some of them, such as the 5 minor deviations and 2 major deviations on D_2cm [29].

By designing a comprehensive comparison between original Type-B plan, the re-calculated Type-C plan, the re-normalized Type-C plan, and the re-optimized Type-C plan, we have clearly outlined in our study the dosimetric necessity as well as impact of re-calculation, re-normalization, and re-optimization using Type-C algorithms. While Li et al.’s study also considered re-optimization using the Type-C algorithm, the conventional conformal beam treatment technique was used in their study, and their re-optimization involved only adjusting the relative beam weights while keeping all other beam parameters the same as the original Type-B plans [29]. Furthermore, we performed our investigation on a much larger patient cohort compared with the above three studies.

Of our 52 patients, 11 (21.2%) resulted in worsened RTOG dosimetric criteria compliance on the re-normalized Type-C plans and hence necessitated re-optimization with the Type-C algorithm. These patients included all 7 patients with a PTV D_95% reduction over 5%. For the remaining 4 patients, 3 had a PTV D_95% reduction over 4% and 1 had a reduction of 2%. This indicates that for those patients on whom Type-B algorithms will lead to large PTV coverage dose overestimations, optimization by Type-C algorithms will likely be necessary to achieve similar levels of RTOG dosimetric criteria compliance currently achieved by Type-B algorithms. For patients on whom Type-B algorithms will lead to small PTV coverage dose overestimations, plans optimized with Type-B algorithms will likely only need to be re-normalized with Type-C calculation to ensure sufficient and accurate target dose coverage, as well as to achieve similar levels of RTOG compliance. In a small number of these patients, if the RTOG compliance of the Type-B plans are borderline, optimization with Type-C algorithms may also be necessary to achieve the same compliance goal.

For one patient, re-optimization using the Type-C algorithm was unable to fully restore the RTOG dosimetric criteria compliance (Patient #11 in Tables 5, 6, and 7). For this patient, the original plan was fully compliant, and the re-normalized plan resulted in minor deviation in both R_50% and D_2cm. The re-optimized plan restored D_2cm compliance to no deviation, but on R_50% the re-optimized Type-C plan still had a minor deviation. Among our cohort, this patient was also the one with the highest magnitude of PTV D_95% reduction on the re-calculated Type-C plan, with a reduction of 16.7%. This large discrepancy in target dose calculation between Type-C and Type-B algorithms might have partially contributed to the difficulty of re-optimizing a fully-compliant plan with the Type-C algorithm. In addition, the R_50% value of the original Type-B plan for this patient, 3.4, was somewhat close to the minor deviation threshold of 3.8. The re-optimized plan had a value of 4.0, and hence scored as a minor deviation.

For all other patients, re-normalization or re-optimization with the Type-C algorithm achieved similar levels of RTOG compliance for these plans and notably higher target dose heterogeneity as seen in the higher HI values. The higher HI might result partially from the fact that the Type-B algorithm overestimates target peripheral dose but underestimates maximum target dose relative to the Type-C algorithm, and partially from the fact that target dose homogeneity was not heavily constrained during re-optimization. Since the current RTOG protocols recommend a wide range of prescription isodose levels from 60 to 90%, the HI values on the Type-C plans for all patients fell within this range.

Since their introductions into commercial treatment planning systems, many clinical studies have been conducted to compare dose calculation of Type-C algorithms to algorithms of previous generations [22, 23, 25, 31, 44‐47]. However, their utilization in clinic practice is still very limited [5]. Two factors contributing to the lack of clinical utilization that motivated our study were: (1) the large patient-to-patient variation of error magnitudes for the current algorithm left it unclear if it is necessary to clinically switch the algorithm at the cost of increased computational time. (2) the four existing studies on the applicability of current dosimetric guidelines for Type-C algorithms had apparent disagreements in their findings [27, 29, 39, 40]. On the first issue, our study using a large patient cohort provided distribution data of target dose errors associated with Type-B calculation for different patients. Clinicians may use these data to assist making decision whether the algorithm switching is necessary. On the second issue, our study resolved the paradox from the existing literature by demonstrating the difference between re-calculated and re-normalized Type-C plans. In addition, the results on re-optimized plans shed additional light on the Type-C applicability of current RTOG guidelines, and further provided clinicians with information on different levels of Type-C involvement in treatment planning necessary to satisfy the dosimetric constraints.

One limitation of our work was that, although extensively comparing the original Type-B and re-calculated, re-normalized and re-optimized Type-C plans on a large patient cohort and with a rich pool of important dosimetric endpoints, our comparison was conducted using one Type-B algorithm, AAA, and one Type-C algorithm, AXB. In particular, although usually considered as a Type-B algorithm and currently used as a major clinical treatment planning algorithm [19‐21, 28, 48‐50], AAA has been shown to perform less accurately in heterogeneous environment than some other Type-B algorithms such as CCC [18, 28, 30]. However, due to the limitations of the treatment planning system, namely Pinnacle does not allow importing plans from other treatment planning systems for re-calculation and it also does not have a Type-C algorithm, a comparison to additional Type-B and Type-C algorithms could not be performed on our cohort.

It would be an interesting direction for future studies to assess the clinical impact of switching from Type-B to Type-C algorithms for lung SBRT. For example, on a cohort of patients treated with Type-B plans, the local failure rates can be compared between those patients with large PTV D_95% or V_100% reductions upon Type-C re-calculation and those patients with small reductions. This way the clinical significance of the dosimetric differences can be determined. However, due to the relatively small percentage of patients with large dose differences between Type-B and Type-C calculations (for example, in our cohort, 13.5% with a PTV D_95% reduction over 5%, or 34.6% with a PTV V_100% reduction over 5%) and the very high local control rates usually at around 90%, a very large cohort with sufficient longitudinal follow-up would be necessary to power such a study and illustrate the clinical significance of the dosimetric effects.

Type-B dose calculation considerably overestimates target dose coverage in lung SBRT for some patients, necessitating Type-C re-normalization or re-optimization. Current RTOG dosimetric criteria appear to remain appropriate in the setting of Type-C calculations.