Automation and integration of a novel restricted single‐isocenter stereotactic body radiotherapy (a‐RESIST) method for synchronous two lung lesions

Abstract Synchronous treatment of two lung lesions using a single‐isocenter volumetric modulated arc therapy (VMAT) stereotactic body radiation therapy (SBRT) plan can decrease treatment time and reduce the impact of intrafraction motion. However, alignment of both lesions on a single cone beam CT (CBCT) can prove difficult and may lead to setup errors and unacceptable target coverage loss. A Restricted Single‐Isocenter Stereotactic Body Radiotherapy (RESIST) method was created to minimize setup uncertainties and provide treatment delivery flexibility. RESIST utilizes a single‐isocenter placed at patient’s midline and allows both lesions to be planned separately but treated in the same session. Herein is described a process of automation of this novel RESIST method. Automation of RESIST significantly reduced treatment planning time while maintaining the benefits of RESIST. To demonstrate feasibility, ten patients with two lung lesions previously treated with a single‐isocenter clinical VMAT plan were replanned manually with RESIST (m‐RESIST) and with automated RESIST (a‐RESIST). a‐RESIST method automatically sets isocenter, creates beam geometry, chooses appropriate dose calculation algorithms, and performs VMAT optimization using an in‐house trained knowledge‐based planning model for lung SBRT. Both m‐RESIST and a‐RESIST showed lower dose to normal tissues compared to manually planned clinical VMAT although a‐RESIST provided slightly inferior, but still clinically acceptable, dose conformity and gradient indices. However, a‐RESIST significantly reduced the treatment planning time to less than 20 min and provided a higher dose to the lung tumors. The a‐RESIST method provides guidance for inexperienced planners by standardizing beam geometry and plan optimization using DVH estimates. It produces clinically acceptable two lesions VMAT lung SBRT plans efficiently. We have further validated a‐RESIST on phantom measurement and independent pretreatment dose verification of another four selected 2‐lesions lung SBRT patients and implemented clinically. Further development of a‐RESIST for more than two lung lesions and refining this approach for extracranial oligometastastic abdominal/pelvic SBRT, including development of automated simulated collision detection algorithm, merits future investigation.


| INTRODUCTION
Stereotactic body radiation therapy (SBRT) of synchronous multiple primary or metastatic lung lesions can result in excessively long treatment planning and delivery times for patients and busy clinics.
To alleviate this process, a single-isocenter intensity modulated radiation therapy (IMRT) or volumetric modulated arc therapy (VMAT) SBRT plan is a feasible treatment option for patients presenting with synchronous multiple metastatic or primary lung lesions. [1][2][3][4][5] SBRT of two lung lesions with a single-isocenter VMAT plan significantly decreases treatment delivery time, increases patient comfort and compliance, and reduces the chance of intrafraction tumor motion errors. 5 However, small patient setup errors may occur due to the difficulties of multiple lung lesions alignment on a single cone beam CT (CBCT) scan. These small setup errors may lead to unacceptable loss in target(s) coverage due to lung heterogeneities and the steep dose gradients obtained in the SBRT plan. 6 Thus, a Restricted Single-Isocenter Stereotactic Body Radiotherapy (RESIST) method was developed to minimize the problems associated with a singleisocenter VMAT lung SBRT plan (e.g., setup errors, collision issues). 7 It has been reported on a multi-institutional database of 700 patients treated with SBRT that patient outcome is related to a clinic's experience in delivery of SBRT. 8 There are no definitive treatment planning guidelines for inexperienced clinics in the treatment of multiple lesions lung SBRT who wish to treat their patients efficiently and accurately.
Recently, a few investigators have presented their work on the use of automation for generating lung SBRT treatment plans using a knowledge-based planning (KBP) approach with dose volume histogram (DVH) estimates via RapidPlan (RP) modeling (Varian Medical Systems, Palo Alto CA). [9][10][11] KBP models can generate plans quickly and improve plan quality and consistency by reducing interplanner variability. These models are trained using previously treated high quality treatment plans and provide a good starting point for subsequent plan optimization. However, there has yet to be a KBP model to automate treatment planning for multilesion lung SBRT including isocenter placement, deploying beam geometry, assigning appropriate dose calculation algorithm and optimizing the plan. In order to guide planners in generating singleisocenter/multi-lesions VMAT lung SBRT plans, an automated treatment planning routine (a-RESIST) has been developed using the RESIST planning geometry, which is further optimized using an inhouse trained KBP lung SBRT model. 11 In the a-RESIST plan, placement of single-isocenter at the mediastinum avoids potential patient/gantry collisions, provides greater flexibility of noncoplanar partial arcs geometry and eliminates the need for multiple couch movements during CBCT imaging. In between the plans, the therapists do not need to enter the treatment vault to reposition the patient because the a-RESIST plans share the same isocenter and the isocenter placement at patient midline ensuring that the daily CBCT imaging will clear the patient without applying a couch shift (couch shift needed for Varian Linac for off-center patients >5 cm laterally). Thus, a-RESIST reduces the chance of a geometric miss due by allowing daily pretreatment CBCT soft tissue matching of one tumor at a time. Moreover, the physician can choose to treat only one lesion per treatment without causing any error in dose tracking (if needed) to manage the patient for various clinical reasons, such as reducing the lung toxicity or if patient cannot tolerate the entire course of treatment. This report aims to demonstrate the feasibility of the a-RESIST treatment planning technique and its ability to assist planners in improving planning efficiency, consistency, and accuracy. Furthermore, this report also provides guidance for automating treatments and simplified workflow for the therapists for the fast and effective synchronous multiple lesions lung SBRT-potentially allowing for offline adaptive replanning, if required.

2.A | Phantom measurements
First, the independent dose validation was performed using the MD Anderson's SBRT credentialing phantom with two targets (spine and lung) by delivering a SBRT prescription dose of 6.0 Gy to the both targets using a single-isocenter VMAT plan following NRG-BR001 protocol. 1 Distance between the spine and lung targets were about 9 cm apart. All dosimetric criteria established by IROC for SBRT treatments to multilesions using single-isocenter approach were satisfied. Second, for our TrueBeam Linac, to quantify the spatial positioning accuracy of a single-isocenter/multitargets plan as a function of distance from the isocenter, the end-to-end phantom tests were performed. Because of the lack of specialized multitargets phantom in our center, for the end-to-end tests, we utilized clinically available catphan phantom with multiple imaging inserts at the different planes and the MPC phantom with 16 bearing balls (BBs) at the different locations. On these phantoms' measurements, it has been observed that our CBCT based target localization accuracy on our Truebeam Linac was within 1 mm (average, 0.75 mm) at 7 cm and <1.2 mm (average, 0.81 mm) at 10 cm distance from the isocenter respectively.

2.B | Patient CT simulation and contouring
After obtaining institutional review board approval, patients were retrospectively selected with two lung tumors each who were previously treated to 50 Gy in 5 fractions using a single-isocenter VMAT CRITCHFIELD ET AL. | 57 lung SBRT following RTOG guidelines. 12 For each patient, both lesions were treated at the same time every other day. All patients were immobilized with the Body Pro-Lok TM Table 1 summarizes the tumor characteristics and tumor distance to isocenter for the ten multilesions lung SBRT patients included in this study. Four lesions were within 2 cm distance from the principal bronchial tree. Distance to isocenter was calculated by finding the Cartesian coordinates of the each PTV geometric center and determining the Euclidian 3D distance with the isocenter coordinates for each plan.
Due to the limited field-of-view (2-3 cm superior to inferior direction for the tumor location) of our 4D-CT scan, our clinical treatment plans were generated on the free breathing planning CT images.

2.C | Clinical VMAT plans
All these patients were treated using a clinical single-isocenter lung SBRT plan that was generated in Eclipse TPS using a Truebeam Linac (Varian Medical Systems, Palo Alto, CA) with the Millennium 120 MLC. All VMAT plans were generated manually utilizing 6 MV-FFF (1400 MU/min) beams. The isocenter was placed approximately between the two tumors. For patients who presented with bilateral tumors or select unilateral tumors, two to three full co-planner arcs were used for treatment. For the remaining unilateral cases, three to five partial co-planner or noncoplanar arcs with couch rotations up to ± 10°were utilized (planner preference). Collimator angles were manually chosen to reduce the MLC leakage dose between each arc with the jaw-tracking feature enabled. 15 Dose was 50 Gy in 5 fractions for all patients. Target naming convention (PTV1 or PTV2) was chosen by the treating physician. Both PTVs were planned with dose prescribed to the 70-80% isodose lines and optimized such that at least 95% of each PTV received 100% of the prescription dose. The maximum dose to each target was planned to fall inside the GTV.
Dose was calculated using the Boltzmann transport based AcurosXB algorithm in Eclipse with heterogeneity corrections with a 1.25 mm calculation grid size (CGS). 13 Reporting dose to medium and photon optimizer (PO) MLC algorithm was used. Although single-isocenter SBRT was designed for synchronous treatment of two lesions, planning objectives per RTOG protocols and NRG-BR001 were utilized for the organs-at-risk (OAR). 1,12,14 Each patient was treated every other day with the VMAT planning technique using an in-house CBCT-guided lung SBRT protocol.

2.D | m-RESIST VMAT plans
Each patient's clinical treatment plan was replanned using the manual RESIST (m-RESIST) method. The m-RESIST method places isocenter at the patient's midline and both tumors share the treatment isocenter. If the lesions are separated in the x-direction, the isocenter is placed approximately between the lesions in the mid-coronal plane of the patient. A separate plan is made for PTV1 and PTV2.
Each plan has three partial noncoplanar VMAT arcs with a 6 MV FFF (1400 MU/min) beam deployed on the tumor side of the patient. Couch rotations were 0°, 10°, and 350°for each beam respectively. Collimator angles were offset by 30°to reduce leakage dose in the same plane and were chosen to ensure that the MLCs can travel to the PTV locations. The new aperture shape controller (ACS) feature in the PO MLC algorithm was set to "very high" in order to reduce the total number of monitor units, reduce plan complexity, and improve plan quality as demonstrated by the previous researchers. 16,17 Briefly, m-RESIST plans were created by fitting the MLCs to PTV1 and then calculating the dynamic conformal arc (DCA) dose for the respective plan. Next, standard manual VMAT T A B L E 1 Main tumor characteristics of the ten lung SBRT patients included in this study. Each patient had two tumors. STD = standard deviation.

Parameters and plans
Mean ± STD (range or n = no. of patients) optimization began for PTV1 and GTV1 coverage. The jaw tracking option was employed to reduce leakage dose to normal lung as described above. 16 Once dose calculation was complete, the plan for PTV1 was used as a base-dose plan before VMAT optimization in the plan for PTV2. The PTV2 plan was optimized for coverage to PTV2 and GTV2 and to spare the OAR. Once optimized and calculated, a m-RESIST plan summation was created with both plans and re-normalized to account for contribution from each plan. The plans were then evaluated per lung SBRT protocols.

2.E | a-RESIST VMAT Plans
Varian Eclipse Scripting Application Programming Interface (ESAPI, Version 15.5) allows for integration of writable scripts and supports the automation of SBRT plans. 18

5.
Offset collimator angles based on PTVs distance to isocenter to allow for optimal MLC travel distance to PTVs.

7.
Application of normal tissue objective and jaw tracking to be used in optimization.   Table 2). This dose escalation is desirable in SBRT treatment since the normal tissue dosing was still acceptable.
For instance, the mean GTV dose for a-RESIST was 7% and 6% higher (up to 3.5 Gy) compared to m-RESIST and clinical VMAT plans respectively.

3.B | Treatment planning parameters
The average total treatment planning time for the a-RESIST script to complete all ten lung SBRT patients with two lesions was 12.5 ± 3.5 min (9.1-21.1 min). Time was recorded on average of 66 min for m-RESIST plans to complete the same tasks as a-RESIST.

3.C | Treatment delivery parameters
The total number of monitor units for m-RESIST and a-RESIST is about 1.8 times higher than for the clinical VMAT plans, as can be seen in Table 3. However, due to both PTVs being planned separately with separate prescriptions, the average modulation factor for the RESIST methods are lower compared to the clinical VMAT method, could potentially improve treatment delivery accuracy. The  an individual plan which are then evaluated with a plan summation.
Allowing each lesion to be planned individually allows for optimal collimator angles and the best use of the jaw tracking feature to aid in the reduction of the normal lung dose. Furthermore, two plans sharing the same isocenter allows for more flexibility during patient treatment as demonstrated in Fig. 5  F I G . 5. Demonstrated is the a-RESIST treatment delivery workflow for a singleisocenter/two-lesion VMAT lung SBRT. The physician has the opportunity to match one lesion at a time and treat without entering the room to re-setup the patient for the second CBCT thus improving treatment delivery efficiency and accuracy by reducing the chance of a geometric miss. Placement of an isocenter at the mediastinum avoids potential patient/gantry collisions, provides greater flexibility of non-coplanar arc geometry, and eliminates the need for multiple couch movements during CBCT imaging. Further improvement of a-RESIST is ongoing in our center including improvement of the KBP optimization model for two-lesion lung SBRT plans and standardizing a more "patient-specific" approach to isocenter placement that could minimize tumor distance to isocenter, while still keeping the patient's midline isocenter. Simulated collision detection is a feature available when using Varian HyperArc module for intracranial SRS treatments and has been further developed by the multiple researchers. [30][31][32] However, simulated collision detection for extracranial SBRT has yet to be studied and would be the next step to the a-RESIST method to further ensure an efficient treatment delivery by automatic collision detection, further reducing overall treatment time. Efficacy of a-RESIST has been demonstrated for two synchronous lung lesions SBRT that could potentially allow for offline adaptive replanning (if required) and can potentially be used for more than two lung lesions as well as other extracranial treatment sites, such as multilesion liver SBRT or oligometastastic abdominal/pelvic lymph nodes SBRT.

| CONCLUSION
Using the a-RESIST planning method for synchronous lung lesions can significantly decrease treatment planning time (<20 min) and allow planners to create clinically acceptable lung SBRT plans. The RESIST method reduces the chance of a geometric miss due to setup uncertainties by allowing for planning and setup of each lesion individually, permitting tumor-to-tumor matching on daily CBCT. Furthermore, automation of the planning technique will allow for standardized treatment plans while allowing user input to further increase the plan quality and treatment efficiency. Utilizing an inhouse trained lung SBRT RP model helps ensure that treatment plans are of consistent high quality. Further improvement of the a-RESIST script to ensure more precise patient specific isocenter placement algorithm as well as well-trained KBP models for patient-specific multitargets lung SBRT that could further improve plan quality, reduce inter-planner variability and inconsistency, and improve patient safety and clinic workflow-potentially allowing for offline adaptive replanning is desired.

CONFLI CT OF INTEREST
The authors declare no conflict of interest.

FUNDING
None.

AUTHOR' S CONTRIBUTION
DP and LCC designed the project. LCC wrote the a-RESIST script, collected and analyzed the data. DP, RM, MR, and MB provided clinical expertise and supervision of the paper. JV and DP developed the lung SBRT RapidPlan model. LCC and DP drafted the manuscript and all co-authors revised and approved the final manuscript.

D A T A A V A I L A B I L I T Y S T A T E M E N T
Research data are not shared.