Clinical impact of the VOLO optimizer on treatment plan quality and clinical treatment efficiency for CyberKnife

Abstract With the recent CyberKnife treatment planning system (TPS) upgrade from Precision 1.0 to Precision 2.0, the new VOLO optimizer was released for plan optimization. The VOLO optimizer sought to overcome some of the limitations seen with the Sequential optimizer from previous TPS versions. The purpose of this study was to investigate the clinical impact of the VOLO optimizer on treatment plan quality and clinical treatment efficiency as compared to the Sequential optimizer. Treatment plan quality was evaluated in four categories of patients: Brain Simple (BS), Brain Complex (BC), Spine Complex (SC), and Prostate (PC). A total of 60 treatment plans were compared using both the Sequential and VOLO optimizers with Iris and MLC collimation with the same clinical constraints. Metrics evaluated included estimated treatment time, monitor units (MUs) delivered, conformity index (CI), and gradient index (GI). Furthermore, the clinical impact of the VOLO optimizer was evaluated through statistical analysis of the patient population treated during the 4 months before (n = 297) and 4 months after (n = 285) VOLO introduction. Significant MU and time reductions were observed for all four categories planned. MU reduction ranged from −14% (BS Iris) to −52% (BC MLC), and time reduction ranged from −11% (BS Iris) to −22% (BC MLC). The statistical analysis of patient population before and after VOLO introduction for patients using 6D Skull tracking with fixed cone, 6D Skull tracking with Iris, and Xsight Spine tracking with Iris were −4.6%, −22.2%, and −17.8% for treatment time reduction, −1.1%, −22.0%, and −28.4% for beam reduction and −3.2%, −21.8%, and −28.1% for MU reduction, respectively. The VOLO optimizer maintains or improves the plan quality while decreases the plan complexity and improves treatment efficiency. We anticipate an increase in patient throughput with the introduction of the VOLO optimizer.


| INTRODUCTION
The CyberKnife robotic system is specifically designed for stereotactic radiosurgery (SRS) and stereotactic body radiosurgery (SBRT). 1,2 The most recent CyberKnife (Model M6 TM ) is equipped with three collimators: fixed cone, Iris TM variable collimator, and InCise TM multileaf collimator (MLC).3,4 Both fixed cone and Iris have 12 circular collimator sizes ranging from 5 mm to 60 mm. The Iris collimator was designed to emulate the fixed cone but with better efficiency in delivery by freely changing collimator sizes at the same delivery position. 5 The goal of the treatment planning process is to optimize the beam aperture, beam angle, and beam weight to achieve good conformity and a steep dose gradient around the target volume, and to minimize the dose to the nearby critical structures. As the treatment targets are often located near or adjacent to the critical structures, the optimization process is often a tradeoff process between the multiple critical goals of target coverage and critical structure sparing.
In 2008, Schlaefer et al. introduced a stepwise optimization algorithm with the approach of optimizing multiple clinical goals in steps with built-in priority. 6 It was implemented in the CyberKnife planning system as the Sequential optimizer. The system searches for a solution under a set of constraints that must be met, and then optimizes the clinical objectives in sequential steps with higher priority for the top objectives. The Sequential optimizer is relatively efficient for simple cases while it shows significant weaknesses on complicated cases. For example, the optimizer seldom converges on the optimal solution when planning with dose escalation or when planning with multiple targets at different dose prescriptions. Furthermore, delivery efficiency is not integrated in the optimization. Therefore, a separate time and beam reduction process has to be performed after optimization. For plans using MLCs, the Sequential optimizer optimizes on pre-created shapes which makes the planning with MLC collimator significantly difficult and inefficient. Finally, long optimization times are needed for complex cases.
To overcome some of the weaknesses with the Sequential optimizer, Accuray (Sunnyvale, CA, USA) released an upgrade to their treatment planning software (TPS) for CyberKnife treatments in November 2018. The upgrade (Precision 1.0 to Precision 2.0) included the VOLO optimizer which was a major rework of the optimization engine used in the TPS and was intended to improve on optimization performance (faster optimization and better plan quality with shorter treatment times), ease of use (intuitive interface with optimization approach similar to other planning systems), and better integration (no time reduction tools required and all plans are deliverable after final calculation).
The VOLO optimizer combines dose-volume histogram (DVH) goals into a single cost function. The goal's importance is specified as objective weighting. For circular collimators (fixed cones and Iris collimation), plan optimization is single phased with direct beam optimization before final dose calculation. The optimization is based on pregenerated beams as the previous optimizer. For MLC collimation, plan optimization consists of two phases: (a) fluence optimization followed by (c) segmentation and aperture adaptation before final dose calculation. The upgraded MLC optimization workflow also includes an interactive DVH display that allows for parameter adjustment during the optimization process. The optimization process integrates delivery efficiency as part of the cost function, resulting in an always deliverable plan. This is in contrast to the Sequential optimizer.
The goal of this study was to investigate the clinical impact of the VOLO optimizer in terms of machine performance, patient throughput, and treatment plan quality compared to the Sequential optimizer.

2.A | Patient selection
Patients who had previously undergone CyberKnife treatment at Stanford Cancer Institute using the Sequential optimizer of Precision 1.0 on the M6 CyberKnife system (Accuray, Sunnyvale, CA) with either Iris 3 or InCise 2 MLC 4 collimation were included. Patients treated with plans using fixed cones were not included as fixed cones are generally only used for small metastatic brain lesions (< 3.0 cm 3 ), for which we expected minimal differences between the optimizers. Five patients were selected from the predefined categories of Brain Simple (BS), Brain Complex (BC), Spine Complex (SC), and Prostate (PC) (Fig. 1). Care was taken to ensure that the plans were representative of the cases typically seen in the clinical practice at Stanford Cancer Institute.

2.B | Treatment plan optimization
Treatment plans previously generated with the Sequential optimizer prior to the TPS upgrade were re-optimized with the VOLO optimizer. Both the BC and the SC cases all had previously generated plans using both Iris and MLC for the purpose of comparison. One of the plans was used for treatment. During the re-optimization process, the prescription dose, fractionation schedule, coverage (volume of tumor that receives ≥ prescription dose divided by the total tumor volume), and maximum dose were kept constant between the plans. The treatment time was not kept constant but was evaluated after optimization to ensure that it was kept within clinically appropriate delivery times. BS cases were optimized using only Iris collimation as these cases would never have been considered for MLC collimation due to the small size and simplicity of the lesions. In contrast, the PC cases were optimized using only MLC collimation due to the size and complexity of prostate treatments. Both BC and SC cases were re-optimized using both Iris and MLC collimation with VOLO optimizer. All treatment plans were generated by medical physicists with >10 years' experience in CyberKnife treatment planning. All plans were deemed clinically acceptable after critical review by physicians and satisfied the dose constraints proposed in TG-101. 7 All final doses were calculated using RayTracing dose calculation algorithm, although Monte Carlo algorithm is also available GI was defined as the ratio of volume receiving ≥ 50% of prescription dose to volume receiving 8 One MU is equal to 1 cGy of absorbed dose in water under calibration conditions (depth = d max , Source-Axis Distance (SAD) = 80 cm, field size = 60 mm diameter at SAD = 80 cm).

2.D | Population evaluation
Two population analyses of patients treated before and after the

2.E | Statistical analysis
The statistical analyses of the population evaluations were performed using unpaired t-tests.  Fig. 2). The CI was 1.16 and 1.14 (1% reduction) and the GI was 3.25 and 3.00 (8% reduction) with the Sequential and VOLO optimizers, respectively. Figure 2 shows a representative case in BS category. This particular plan did not show a significant reduction in treatment time  with the VOLO compared to the Sequential optimizer, but a reduced dose gradient and an increased conformity was shown.

3.B | Brain Complex (BC)
For BC plans using Iris collimation, the average treatment time and

3.E | Population Analysis
All patients planned and treated on our CyberKnife system during the 4 months before and after TPS upgrade were compared (Fig. 6).
A significant improvement in plan efficiency was found for all metrics for plans generated with Iris collimation regardless of site. The number of beams and number of MUs was reduced by~22% and~28%, respectively, for both brain and spine. The corresponding values for average treatment time reduction was 22.2% (7 min reduction) and 17.8% (8 min reduction), respectively. No significant difference was found in brain cases using fixed cone collimation. The number of plans with MLC collimation for both tracking types, and the number of plans using Fixed Cone collimation for Xsight Spine tracking were too few for analysis.
The collimation usage was analyzed on a monthly basis pre-and post-TPS upgrade (Fig. 7). For brain cases, increased consideration for MLC usage was seen for more complex cases of brain lesions.
However, the distribution between fixed cone and Iris collimation We did not include such plans in our designed comparison due to the difficulty in quantitative CI and GI calculations. To illustrate this improvement, one re-plan with two targets at different dose level was presented in Figure S1. included, the OAR doses were intentionally left out in the results.
For example, the brain cases were chosen based on treatment complexity, not on similarity between cases. Between the different plans, not only the relevant risk organs differ, but also the fractionation scheme and total dose. However, in the case of spine and prostate, the same type of risk organ is present in all plans. Therefore, we have included the risk organ doses for these two categories in Table S1 and Table S2.  Due to that in the majority of cases, the Iris plans outperformed the MLC plans using the Sequential optimizer in Precision 1.0, too few plans were available in the population analysis for proper statistical handling. However, in the comparison of the re-optimized plans, a marked improvement in plan quality was seen, with, for example, significant improvements in conformity in BC and SC categories for the MLC plans. With these data, MLC has seen increased consideration of use in brain and spine cases after the TPS upgrade. A real increase in the MLC usage was not seen until 4 months after the upgrade due to MLC hardware and software compatibility issues.
Once this was resolved (at 4 months) the MLC usage has steadily increased, especially in cases related to the spine (Fig. 7).
With the significant reduction of MU and treatment time for plans using VOLO optimizer, the other potential benefits may be the reduced patient body dose from the head and collimator leakage, and decreased imaging dose. While patient body dose is directly proportional to MU, the imaging dose is proportional to treatment time.
A pair of KV images is usually taken at an interval between 30 amd 90 sec based on patient positioning stability.

| CONCLUSION S
The VOLO optimizer maintains or improves the plan quality while decreases the complexity compared to the Sequential optimizer with the introduction of the VOLO optimizer an increase in MLC collimation usage was seen. For plans generated with Xsight Spine tracking, the MLC usage was higher than Iris usage at five and six months after the TPS upgrade.