Assessment of multi‐criteria optimization (MCO) for volumetric modulated arc therapy (VMAT) in hippocampal avoidance whole brain radiation therapy (HA‐WBRT)

Abstract This study compared the dosimetric performance of (a) volumetric modulated arc therapy (VMAT) with standard optimization (STD) and (b) multi‐criteria optimization (MCO) to (c) intensity modulated radiation therapy (IMRT) with MCO for hippocampal avoidance whole brain radiation therapy (HA‐WBRT) in RayStation treatment planning system (TPS). Ten HA‐WBRT patients previously treated with MCO‐IMRT or MCO‐VMAT on an Elekta Infinity accelerator with Agility multileaf collimators (5‐mm leaves) were re‐planned for the other two modalities. All patients received 30 Gy in 15 fractions to the planning target volume (PTV), namely, PTV30 expanded with a 2‐mm margin from the whole brain excluding hippocampus with margin. The patients all had metastatic lesions (up to 12) of variable sizes and proximity to the hippocampus, treated with an additional 7.5 Gy from a simultaneous integrated boost (SIB) to PTV37.5. The IMRT plans used eight to eleven non‐coplanar fields, whereas the VMAT plans used two coplanar full arcs and a vertex half arc. The averaged target coverage, dose to organs‐at‐risk (OARs) and monitor unit provided by the three modalities were compared, and a Wilcoxon signed‐rank test was performed. MCO‐VMAT provided statistically significant reduction of D100 of hippocampus compared to STD‐VMAT, and Dmax of cochleas compared to MCO‐IMRT. With statistical significance, MCO‐VMAT improved V30 of PTV30 by 14.2% and 4.8%, respectively, compared to MCO‐IMRT and STD‐VMAT. It also raised D95 of PTV37.5 by 0.4 Gy compared to both MCO‐IMRT and STD‐VMAT. Improved plan quality parameters such as a decrease in overall plan Dmax and total monitor units (MU) were also observed for MCO‐VMAT. MCO‐VMAT is found to be the optimal modality for HA‐WBRT in terms of PTV coverage, OAR sparing and delivery efficiency, compared to MCO‐IMRT or STD‐VMAT.


| INTRODUCTION
Brain metastases are an important source of morbidity for cancer patients. Whole brain radiation therapy (WBRT) is effective, but results in significant neurocognitive side effects for many patients, especially in terms of verbal memory. As survival for patients with metastatic brain disease increases, 1,2 approaches to spare neurocognition have become an intense area of study. Focal radiation with stereotactic radiosurgery (SRS) is one approach that results in less neurocognitive impairment, 3 but is not an option for many patients with more diffuse metastatic disease. One alternative that has gained popularity in the last several years has been hippocampal avoidance whole brain radiation therapy (HA-WBRT), which uses advanced radiation techniques to reduce the dose to the hippocampus, an area important for memory formation and neurogenesis. 4 The RTOG 0933 phase II study showed evidence of improvements in quality of life and memory preservation compared to historical WBRT controls. 4 Hopkins Verbal Learning Test-Revised Delayed Recall (HVLT-R) revealed a 30% mean relative decline in WBRT without hippocampal avoidance (baseline 4 months) versus 7% utilizing HA-WBRT along with no decline in Quality of Life scores (QOL). 4 Intensity modulated radiation therapy (IMRT) has been used as a practical delivery method for HA-WBRT based on RTOG 0933 guidelines. 5 Dose painting to metastatic lesions, although not required by the RTOG protocol, has also been examined. 6 Despite these efforts, recent survey results from Slade et al. 7 indicated 56% of radiation oncologists (n = 196) would not consider (IMRT) for HA-WBRT; among several factors was the complexity of the treatment planning process which requires substantial training. More recently, volumetric modulated arc therapy (VMAT) has also been examined for HA-WBRT. 8 VMAT showed superior dosimetric performance to IMRT, 9,10 and can practically deliver dose painting in form of simultaneous integrated boost (SIB) 11 to multiple brain metastases, 12,13 along with HA-WBRT. 14,15 Dosimetric quality and efficiency for IMRT and VMAT were further promoted by a recent advancement in inverse planning technology: multi-criteria optimization (MCO). [16][17][18][19][20][21] MCO generates a Pareto surface containing a spectrum of optimal plans, with every point on the surface representing an optimal solution with different trade-off objectives. 16 A user is able to navigate combinations in real-time based on specified trade-off objectives along with planning constraints. 16 Numerous studies have confirmed that MCO improved plan quality over conventional inverse planning methods. [17][18][19][20][21] In addition, MCO also reduced planning time and allowed less-experienced treatment planners to efficiently produce highquality IMRT plans for complex targets in the close vicinity of numerous organs-at-risk (OARs), such as tumors in the head and neck region. 20 The study aimed to compare three treatment planning methods - The effectiveness of using SIB to deliver the extra dose to the metastatic lesions was also assessed for all three methods.

2.A | Patient selection
Ten patients previously treated with HA-WBRT using MCO-IMRT

2.B | Computed tomography simulation
Computed tomography (CT) data were originally prepared based on the RTOG 0933 criteria. All patients had MRI with axial T2-weighted and gadolinium contrast-enhanced T1-weighted sequences for hippocampus contouring with slice thickness no greater than 1.5 mm. They were immobilized in the supine position with a thermoplastic mask for a CT simulation with slice thickness of 1.25 mm with intravenous contrast.
The MRI images were semi-automatically fused to the simulation CT by an attending radiation oncologist in MIM Vista version 6 (MIM Software Inc., Cleveland, OH). Target structures (such as whole brain and distinguishable metastatic lesions), organs-at-risk (OARs) and external patient contour w/immobilization devices were also contoured within MIM before exporting to the treatment planning system (TPS). The hippocampus was contoured based on the RTOG contouring guidelines.
The hippocampal avoidance region was generated by a 5 mm contour expansion followed by a secondary 5 mm expansion to control the dose gradient in the avoidance region. There were up to two levels of planning target volume (PTV). The hippocampal sparing whole brain PTV included the whole brain parenchyma to C1 or C2 as the clinical target volume (CTV) plus 2 mm expansion with subtraction of the hippocampal avoidance regions. The metastatic PTV was expanded from the delineated metastatic lesions with a 2 mm expansion.

2.C | Treatment prescription
Ten patients received 30 Gy in 15 fractions to the hippocampal sparing whole brain PTV (PTV30) with an SIB of 37.5 Gy to specific metastatic PTV (PTV37.5).

2.D | Custom optimization contours
All plans were given custom contours to guide the optimization process. These contours included PTV30ÀPTV37.5, a volume used to ZIEMINSKI ET AL.
| 185 help improve dose uniformity within the whole brain region without metastatic disease. A custom target structure was created within PTV30 in between the left and right hippocampus for added control of midline target coverage. A custom avoidance structure was created inferior to the PTV30 to control excessive inferior dose to uninvolved optic regions and oral cavity.

2.E | Treatment plan parameters
MCO-IMRT plans utilized a step-and-shoot delivery method (SMLC) with eight to eleven beams at variable gantry angles (including one to three non-coplanar angles) depending on the location of the metastatic disease. All plans were optimized using a 6-MV photon beam on an Elekta Infinity linear accelerator employing Agility multileaf collimator (MLC, 80 pairs of 5-mm leaves).
All VMAT plans were generated using dual coplanar full arcs The use of multiple arcs has been shown to yield superior plan quality. 22 Furthermore, with the dual arc feature enabled, only one set of fluence profiles are optimized while more information from the fluence maps can be kept during the leaf sequencing process.
This 'one arc' fluence map conserves leaf motion with one arc focusing on the left side of the target and the second arc on the right side at a given control point angle, which in turn reduces the leaf openings over the OAR and increases sparing. 23 Chen et al. reported that the use of dual arcs in VMAT resulted in notable dosimetric improvements for complex targets such as head and neck. 24 The differences between the STD-VMAT (a.k.a., rayArc) and MCO-VMAT optimization in RayStation were explained by Ghandour et al. 25 The rayArc uses a direct machine parameter optimization (DMPO) algorithm that starts with a coarse arc segmentation (24°s pacing), while converting to optimized fluence maps per initial angle.
The maps are then converted into a user determined 2-4 control points per initial angle, while filtering out the smallest points based on a sorting algorithm. 23 Control points are then converted to comply with machine parameter motion constraints (e.g., max leaf speed, valid dose rates, delivery time, number of monitor units per degree) along with leaf/jaw positioning (static or dynamic). Chen et al. reported the use of small arc spacing of 2°per control point led to notable dosimetric improvements for complex targets such as head and neck. 24 MCO empowers the user to produce a final plan by considering multiple criteria via the generation of a Pareto database followed by a navigation process which smoothly interpolates amongst the plans in the database. For a Pareto optimal plan, no criterion can be further improved without sacrificing another criterion. For photon optimization in RayStation, for a plan with n objectives defined, a minimum of 2n Pareto plans are needed to produce a practical approximation of the Pareto surface, 18 whereas the use of 4n plans leads to a closer approximation to the true Pareto surface, which we use at our institute. 19 In addition to objectives, constraints are used to focus on clinically useful plans, creating a Pareto surface with smaller range but finer resolution. At least two constraintsthe minimum dose for a target and the maximum dose for an organ, are required to start the Pareto plan generation. An MCO plan is selected by navigating on the Pareto surface using the navigational sliders. A particular slider can be clamped to limit the range of navigation on the Pareto surface to prevent degradation while navigating other sliders in desired directions. Once navigated, the fluence pattern of the selected Pareto optimal plan is converted to deliverable machine parameters for each control point using a final dose calculation. In RayStation, this final deliverable plan is used for clinical evaluation.

2.F.3 | Plan quality parameters
Total monitor units and Dmax (Gy) for each plan was recorded. Dose uniformity with control of the dose falloff was assessed using the volume of PTV30ÀPTV37.5. The V35 of PTV30 was also recorded to determine the control of dose beyond 30 Gy in the whole brain regions without metastatic disease.

2.F.4 | Statistical analysis
Statistical analysis was performed with a Wilcoxon signed-rank test to determine if there was any significant difference of the parameters examined along with standard deviations (SD). A P value smaller than 0.05 was considered statistically significant. The comparison was conducted between MCO-VMAT and MCO-IMRT, as well as between MCO-VMAT and STD-VMAT. STD-VMAT was not compared to MCO-IMRT. Table 1 shows the total plan MU, dose to OARs, maximum plan dose, and target coverage for the ten patients. The number of meta-

3.A | OAR sparing
As shown in Table 2

3.B | Target coverage
As shown in Table 2 Table 1). Figure 2 shows the D99 of GTV37.5 and D95 for PTV37.5. In general, the three modalities provided similar target coverage considering the challenges of multiple lesions. Figure

| DISCUSSION
All MCO-VMAT plans (with or without the SIB to the metastatic lesions) achieved the RTOG 0933 guidelines (which only required WBRT to 30 Gy) with acceptable or better hippocampus sparing. 26 Prior studies have highlighted MCO for its operational flexibility and planning efficiency, along with superior dosimetric performance. 21,24 T A B L E 2 Cumulative average and standard deviation of the plan metrics shown in Table 1, for each treatment modality. The impact of using VMAT over IMRT in MCO planning is shown under the MCO column, whereas the impact of using MCO over standard optimization in VMAT planning is shown in the column of VMAT. The differences with statistical significance (P < 0.05) is shown in red.

MCO-VMAT
MCO-IMRT STD-VMAT F I G . 1. Isodose plan comparison in the coronal view for the patient that MCO-VMAT demonstrated most improvement on the coverage of the whole brain PTV30 (patient 3 in Table 1). Prokic et al. reported that the SIB technique could achieve better hippocampal sparing compared to sequential boost in form of stereotactic radiation therapy (e.g., 8 Gy 9 2) after WBRT. 28 In addition, the use of a single isocenter for the treatment of multiple brain metastasis led to reduced delivery time while maintaining the dosimetry quality, as compared to the sequential boost. 29

ACKNOWLEDG MENTS
We wish to thank David Craft and Thomas Bortfeld for sharing their valuable feedback and expertise.

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