Introduction
An estimated 13,000 new cases of laryngeal cancer were diagnosed in the United States in 2016 alone, making it the most common non-cutaneous head and neck malignancy [
1]. Most laryngeal cancers occur around the true vocal cord and the glottis larynx region, and most are detected at early stages and can be cured by mono-modality treatments. Radiation therapy, as a local therapy technique, has been highly effective in treating early-stage laryngeal cancer, with local control rates around 90% for Tis and T1 stage tumors, and over 70% for T2 stage tumors [
2]. However, conventional radiation therapy treatments for laryngeal cancers usually use opposed-lateral or wedged-pair beams, which result in considerable high dose irradiation to normal tissues. Population-based studies have suggested an increase in late risk of ischemic events following radiation therapy of head and neck cancers [
3,
4], presumably from high dose irradiation of carotid arteries. Radiation therapy as a treatment for laryngeal cancer is also facing competition from other surgical alternatives, such as trans-oral laser excision, which removes the gross disease and often preserves adjacent portions of the laryngeal skeleton and mucosa. In addition, current larynx radiation therapy normally requires a long treatment course of 30–33 fractions spanning 5–6 weeks, which can be costly and inconvenient. A hypofractionation scheme would reduce the number of fractions to lower the treatment cost and improve patient convenience. Some previous studies have investigated the feasibility of reducing the total number of fractions of larynx radiotherapy to 25–28, and they have achieved equivalent or even better local control rates without increasing toxicity [
5,
6]. These promising results suggest further hypofractionation may achieve better local control and help to minimize the treatment length and cost.
Stereotactic body radiation therapy (SBRT) requires substantially fewer treatment fractions than conventional radiotherapy. Using advanced image guidance and motion control for margin reduction, SBRT may also lower the dose to nearby organs-at-risk (OARs) while simultaneously increasing dose potency to tumors. Thus, SBRT presents a possible solution to the challenges of improving laryngeal cancer treatment with radiation therapy. Our institution recently performed a phase I study of larynx SBRT using the Cyberknife system (Accuray, Sunnyvale, CA) [
7,
8]. The study evaluated the feasibility of using SBRT to treat only the involved site of disease plus a 3 mm margin by large-dose, highly-focal radiation fields. The phase I trial has yielded very encouraging results, with local control rates as good as conventional therapy (> = 80%) [
8]. We chose the robotic Cyberknife system [
9] to deliver radiotherapy by considering several advantages, such as near real-time target tracking capability (offered by the orthogonal x-ray imaging system) and improved dose conformity from non-coplanar beam delivery [
10].
Motivated by the encouraging results of the phase I trial, we recently initiated a phase II trial to further evaluate the efficacy of the larynx SBRT technique. Before conducting this trial, we explored the potential of using a gantry-based LINAC systems as an alternative treatment platform for larynx SBRT. Such LINACs are more widely accessible and can benefit a much larger patient population than the Cyberknife. LINAC can provide 3D visualization of soft tissues from CBCT imaging for localization and set up. Some LINACs also have near real-time imaging/localization capacity via technologies like MV cine imaging, ExacTrac [
11] (BrainLAB AG, Heimstetten, Germany), AlignRT [
12] (Vision RT Ltd., London, UK), Calypso [
13] (Varian Medical Systems, Palo Alto, CA), MR-LINAC [
14] and etc. All these attributes make a LINAC a potential alternative platform for the phase II study.
In this study, we investigated the feasibility and quality of larynx SBRT planning on a conventional gantry-based LINAC. We proposed to use non-coplanar volumetric-modulated arc therapy (VMAT) beams for LINAC treatment planning. Volumetric-modulated arcs [
15] enable more angular coverage than intensity-modulated static beams to preserve planning target volume (PTV) coverage while depositing the dose more uniformly across the whole volume. Using non-coplanar arcs allows more degrees of freedom to approach a large solid angle treatment similar to Cyberknife [
10]. VMAT also provides relatively fast delivery to potentially reduce the effect of intra-fractional target motion on delivered doses. Ten patients with laryngeal cancer treated in our Cyberknife SBRT phase I trial were retrospectively replanned in the Eclipse treatment planning system (TPS) (Varian Medical Systems, Palo Alto, CA) for static dosimetry comparison [
16]. For Cyberknife planning, we used the Monte-Carlo algorithm for dose calculation because of the air cavity within and/or around the treatment volume. For LINAC planning in Eclipse, we used the anisotropic analytical algorithm (AAA [
17]), as it is the most widely used. However, for fair comparison with the Cyberknife plans, we also used the Acuros XB (AXB) algorithm for LINAC plan optimization and dose calculation, because it agrees better with Monte-Carlo than AAA [
18,
19]. We also performed an end-to-end phantom test to validate the accuracy of AXB and AAA for larynx dose calculation. We compared the Cyberknife and LINAC VMAT plans using dosimetric endpoints such as the conformity index (CI) [
20], OAR maximum/mean doses, R50 (ratio of the 50% isodose volume of the prescription dose to the PTV), R20 (ratio of the 20% isodose volume of the prescription dose to the PTV), and the monitor units (MU).
Discussion
Previous studies have found that the AXB algorithm calculates doses at inhomogeneous regions more accurately than the AAA algorithm [
18,
19]. Our findings in the end-to-end phantom measurement study agreed with those studies. Good overall accuracy was observed for either algorithms on a static phantom (Fig.
2), however AXB provided better dose accuracy in the air cavity region. For larynx PTV, the air cavity is included to compensate motion but not for actual dose disposition. We believe that actual dose deposition in soft tissues will be adequate for either AAA or AXB (Fig.
2) with motion. Since AAA reports dose to water [
27], and most clinical outcome studies are based on dose to water [
28,
29], our clinicians are more comfortable interpreting the treatment outcome based on AAA dose calculation. We will continue to use AAA as the dose calculation engine for our phase II trial until we gain more experience with AXB. In the future AXB will likely replace AAA as the main dose calculation engine due to its better accuracy.
Our study shows that larynx SBRT may be planned either on Cyberknife or on conventional LINACs with the same target coverage and similar OAR avoidance (Figs.
3,
4 and
5 and Table
3). There are no statistical differences for most evaluation metrics. For metrics where statistically significant differences were identified, VMAT-AXB and VMAT-AAA plans were either slightly better on selected metrics (i.e., maximum PTV dose, mean thyroid gland dose, maximum contralateral arytenoid dose, and MU), or slightly worse (i.e., R50) than the Cyberknife plans. In contrast, a previous study found worse OAR sparing by LINAC-based coplanar IMRT plans than Cyberknife plans [
7], likely due to the beam arrangement. In our study, the good LINAC plan quality indicates that using VMAT can evenly distribute the beams to achieve better target coverage, and using non-coplanar geometry can further spare OARs. It is worth to mention that the Cyberknife plans offered more compact dose distributions and better conformity indices by using fixed cones [
23]. In contrast, our LINAC used a standard MLC of 5 mm resolution at the isocenter level. A high-definition MLC with a finer leaf width (2.5 mm) may improve the dose fall-off to achieve more compact doses [
30] for our proposed non-coplanar VMAT plans. In general, this comparison study suggested that similar plan quality can be achieved on either LINACs or Cyberknife platforms for larynx SBRT treatments, meeting all the dosimetric constraints on Table
1.
It is found that VMAT plans used less than 1/3 of the total MUs of the Cyberknife plans (Table
3). The large MUs associated with Cyberknife plans are partially due to the small cones used in this study, as MUs and treatment time generally decrease with increasing cone size. However, due to the limited size of the larynx PTV, we found using a small fixed cone necessary to maintain the plan quality, especially on the dose conformity index. Although the MUs can be partially correlated with the net beam-on time, they cannot fully represent the overall treatment delivery time, especially when comparing between two different modalities. The treatment delivery time comparison between LINAC and Cyberknife can be complex, depending on many factors including dose rate, intra-fractional imaging and setup correction. It should be acknowledged that the treatment time, especially for LINAC delivery, can vary among institutions with different intra-fractional verification protocols and technology. Different institutions should compare the treatment delivery time between LINAC and Cyberknife based on their own protocols. In general, shorter treatment time potentially reduces the effect of intra-treatment tumor motion or baseline drift [
31,
32], though Cyberknife systems could be less susceptible to intra-treatment tumor motion from longer treatment time as compared to LINACs, due to its real-time motion compensation strategy. However, it should also be emphasized that shorter treatment time may also indicate higher dose rate, and higher dose rate may potentially increase the incidence of normal tissue toxicity for larynx SBRT from a radiobiological point of view [
33]. Thus careful consideration is warranted in selecting a technology for potential larynx SBRT treatments.
The InCise MLC on the Cyberknife systems may deliver plans more efficiently and help reduce the beam-on time [
25]. Our fixed cone plans were all designed and delivered before our clinic adopted and commissioned an InCise MLC for Cyberknife. Retrospectively, we also performed a preliminary study investigating the potential of using the InCise MLC to plan the CK larynx SBRT cases. For the Cyberknife MLC plans, we used the Cyberknife Precision TPS with the latest VOLO optimizer. Similarly, the Monte-Carlo engine was employed for dose calculation. From the study, we found the MLC plans could not meet all objectives/constraints especially the conformity index constraint. For three evaluated patient cases, the MLC plans yielded conformity index all > 1.4, exceeding our protocol’s hard constraint (1.3). Thus we did not further pursue the use of the Cyberknife MLC to generate the larynx SBRT plans for comparison. Such a discrepancy for MLC plans could be caused by several factors: 1) the small PTV size of larynx plans (2.7 cc – 11.1 cc for this study) for which MLC has found challenging in achieving a good quality plan especially on conformity index [
25]; 2) the “fluence-to-leaf sequence” optimization strategy for MLC, of which the post-optimization leaf sequencing may lead to inferior plan quality; 3) Cyberknife larynx plans have to employ the Monte Carlo dose calculation algorithm due to the air cavity presence. However, the current VOLO optimizer only applies the Monte-Carlo algorithm in the later stages of MLC plan optimization that include segment weighting adjustment and final dose calculation. Instead, it uses pencil-beam-based algorithms for fluence optimization and leaf adaptation, which may lead to the sub-optimal plans after final Monte-Carlo dose calculation of the MLC plans. In contrast, fixed cone plans do not require fluence optimization. And the optimization of fixed cone plans is driven by Monte Carlo dose calculation. As a result, the final dose of fixed cone plans is close to that achieved during optimization. Nonetheless, these limitations on MLC optimization may be overcome with future algorithm updates, and the potential of Cyberknife MLC plans should be re-assessed in the future.
In this comparison study, we used the same margin recipe for both the LINAC VMAT and the Cyberknife plans. In our phase II larynx SBRT trial (NCT03548285, ClinicalTrials.Gov), we are investigating the use of surface imaging-based motion management strategy for LINAC treatments [36]. The skin surface around the larynx region is tracked as a surrogate of intra-fractional tumor motion. Beam-hold is enabled when the surface motion goes beyond a clinically-defined threshold. This strategy allows continuous motion monitoring in real time, potentially achieving similar motion control capability as Cyberknife and hence a similar margin recipe. Therefore, the dose results reported in this study were based on our clinically-realistic plans, and reflected our clinical practices. However, it should be noted that the margin recipes for LINACs or Cyberknife machines might vary among institutions, which are affected by individual technology capability and institutional policy. One may have to weigh in the potential margin differences when generating the LINAC and CK plans for dose comparison, to determine the most appropriate technology to use.
There are some limitations of our study. This planning study is based on static dose calculation which does not account for motion. There are potentially complex interplay effects between the larynx motion, the air cavity and the small, intensity-modulated treatment fields, which may lead to dose deviations from planning. Measurement studies using a motion phantom are warranted to evaluate the effects of larynx motion, and further compare the gantry-based LINAC with Cyberknife for larynx SBRT treatments. Furthermore, the actual delivered dose cannot be easily tracked on patients. In addition to motion-induced dosimetric uncertainty, the differences in dose calculation engines may also introduce dosimetric uncertainties (for instance, AAA vs. AXB), which is difficult to quantify in a patient-specific or organ-specific manner. A further comparison will rely on a future treatment outcome study.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.