Full automation of spinal stereotactic radiosurgery and stereotactic body radiation therapy treatment planning using Varian Eclipse scripting

Abstract The purpose of this feasibility study is to develop a fully automated procedure capable of generating treatment plans with multiple fractionation schemes to improve speed, robustness, and standardization of plan quality. A fully automated script was implemented for spinal stereotactic radiosurgery/stereotactic body radiation therapy (SRS/SBRT) plan generation using Eclipse v15.6 API. The script interface allows multiple dose/fractionation plan requests, planning target volume (PTV) expansions, as well as information regarding distance/overlap between spinal cord and targets to drive decision‐making. For each requested plan, the script creates the course, plans, field arrangements, and automatically optimizes and calculates dose. The script was retrospectively applied to ten computed tomography (CT) scans of previous cervical, thoracic, and lumbar spine SBRT patients. Three plans were generated for each patient — simultaneous integrated boost (SIB) 1800/1600 cGy to gross tumor volume (GTV)/PTV in one fraction; SIB 2700/2100 cGy to GTV/PTV in three fractions; and 3000 cGy to PTV in five fractions. Plan complexity and deliverability patient‐specific quality assurance (QA) was performed using ArcCHECK with an Exradin A16 chamber inserted. Dose objectives were met for all organs at risk (OARs) for each treatment plan. Median target coverage was GTV V100% = 87.3%, clinical target volume (CTV) V100% = 95.7% and PTV V100% = 88.0% for single fraction plans; GTV V100% = 95.6, CTV V100% = 99.6% and PTV V100% = 97.2% for three fraction plans; and GTV V100% = 99.6%, CTV V100% = 99.1% and PTV V100% = 97.2% for five fraction plans. All plans (n = 30) passed patient‐specific QA (>90%) at 2%/2 mm global gamma. A16 chamber dose measured at isocenter agreed with planned dose within 3% for all cases. Automatic planning for spine SRS/SBRT through scripting increases efficiency, standardizes plan quality and approach, and provides a tool for target coverage comparison of different fractionation schemes without the need for additional resources.


| INTRODUCTION
Spinal metastases are devastating complications of malignant tumors that can decrease quality of life due to pain, impaired mobility, or neurologic disturbances. 1,2 Surgical resection has been the mainstay of treatment to reduce tumor burden and stabilize the surrounding anatomy, but may not be feasible in patients with poor performance status or tumors that abut critical organs. 3 Radiation, now widely adopted for the treatment of spinal metastases, can be delivered as adjuvant to surgery, or as definitive treatment, to improve tumor and symptomatic control. The adoption of stereotactic immobilization and real-time image guidance along with advances in treatment planning has allowed for treatment of spinal lesions using larger fraction sizes to improve tumor control and reduce treatment burden. 4 RTOG 0631 phase II concluded that spine SRS was safe and feasible to be implemented in a high-level cooperative group trial. 5 Single institution studies demonstrate an advantage of using SRS/SBRT in terms of increased local control and complete response compared to conventional three-dimensional conformal radiotherapy (3DCRT). [6][7][8][9][10][11][12][13] Sprave et al. 14 reported improved pain response on a randomized phase II trial evaluating single fraction spine SBRT vs 3DCRT. The SABR-COMET randomized phase II trial compared standard 3DCRT palliative radiotherapy to SBRT for oligometastatic patients (including patients with spinal metastases) and reported a median overall survival of 28 months in the control group vs 41 months in the SABR group. Treatment-related deaths occurred in 3 (none of which were spine) of 66 the patients after SABR, compared to none (n = 33) in the control group. 15 The most feared complication from spine SRS is radiation myelopathy. Therefore, the dose to the spinal cord when performing spine SRS needs to be tightly controlled. A retrospective review on 1388 patients reported a 0.4% rate of myelophaty, 16 and prospective studies have presented a range between 0 and 3%. 5,[17][18][19] Other relevant toxicities associated with spine SRS treatments are vertebral compression fractures (VCF) [20][21][22] and esophageal toxicity (fistula, ulcer, stenosis). 23,24 Current approaches to spine SRS/SBRT require generation of complex treatment plans using intensity-modulated radiotherapy

2.A | Subjects
Ten previously treated spine SBRT patients from our institution were anonymized and retrospectively replanned using the automatic treatment planning script. The patient dataset included targets treating the cervical (2), thoracic (7), and lumbar (1) spine regions. CT simulation was performed on a Siemens SOMATOM scanner (Siemens, Erlangen, Germany) with the following scan protocol: 1.5 mm slice thickness, 500 mm acquisition diameter, and extended field of view (FOV) reconstruction of 650 mm. The use of anonymized retrospective CT scans for dosimetry studies was approved by our Internal Review Board (IRB S18-00659).

2.C | Treatment plans
For this study, three plans were automatically generated for each patient according to our radiosensitive de novo prescription scheme

2.D | Patient-specific QA
To evaluate deliverability of the treatment plans, patient-specific QA was performed for all 30 individual plans (ten treatment plans for each fractionation scheme). Dose verification plans were created for a SNC ArcCHECK diode array device and evaluation was carried out using the composite plan. Global gamma metric passing rates were evaluated using three different threshold levels -3% at 2 mm, 2% at 2 mm, and 3% at 1 mm, all with a 10% threshold dose.   Table 3. The beam-on time for delivery was below 6 min for all cases.

| DISCUSSION
Our results demonstrate that it is feasible to establish an automatic workflow to create high-quality treatment plans for spine SRS. The tool presented in this study automated plan creation, optimization, and calculation with the primary goal of meeting all OARs constraints while maximizing target coverage irrespective of lesion T A B L E 1 Dose objectives and results for single fraction and three fraction plans generated automatically for ten cases. location, size, distance, and fractionation. Target coverage varied depending on the need to spare the most critical OARs, such as spinal cord, cauda equina, esophagus, trachea, and brachial plexus that, for some cases, are located adjacent to spinal metastases or even overlap with the PTV. As expected, the target coverage increased when the fractionation was increased due to the increased percent of the prescription dose allowed to be delivered to each OAR. The ability to create different fractionation plans provides an excellent tool to balance target coverage with the most optimal fractionation scheme.
All the plans presented here were obtained with only one opti-  Currently, we are working to incorporate that functionality, allowing the user to override the default dose limit to the spinal cord in terms of maximum dose allowed for the current plan or in terms of a combined biologically equivalent dose (BED). Second, the field setup using four co-planar full arcs and the mentioned collimation rotations, while used in this study and found to provide excellent dose shaping, is not a requirement. Eclipse scripting allows the user to set every single parameter of the field arrangement (isocenter location, couch rotation, collimator rotation, arc rotation), and therefore, the field arrangement can be set automatically as desired using the scripting code. Third, it is important to acknowledge that the treatment plan time presented is an approximation. Fourth, for this study we added together cervical, thoracic, and lumbar spine cases due to the small sample size. The script is able to handle any location and consider any set of OARs contoured for planning. However, different locations present different challenges in terms of planning and influence the overall results reported in this study compared to a stratified analysis based on lesion location. Finally, due the retrospective nature of this project and the limited sample size, the performance of the script and quality of newly generated plans will need to be closely monitored and further validated when used to generate treatment plans for clinical use.
Future developments can be incorporated into our current automatic tool, thanks to the high degree of customization available using scripting. We are currently working on making the tool compatible with re-irradiation treatments. We are also working to incorporate into the script a plan complexity calculation that can be used to evaluate how modulated/complex the plan is even prior to performing patient-specific QA. Furthermore, a knowledge-based planning model is currently in development for incorporation into the automatic optimization step, as previously suggested. Finally, the script can be further customized to employ and compare different beam arrangement configurations than the ones presented in this study.

| CONCLUSION
Automatic treatment planning for spine SRS/SBRT using scripting can facilitate the task of planning for spine metastases. Automatic planning through scripting increases efficiency, standardizes plan quality and approach, and provides a tool for target coverage comparison of different fractionation schemes without the need for additional resources. Median value with range (minimum, maximum) in parentheses. For simultaneous integrated boost (SIB) plans (single and three fractions), the cGy value is based on the highest dose level.