Impact of beam configuration on VMAT plan quality for Pinnacle³Auto-Planning for head and neck cases

The Pinnacle³ Auto-Planning engine (Version 9.10) was used for optimization of VMAT plans. Treatment was planned for an Elekta Synergy Platform® equipped with Agility Head (Elekta Oncology Systems, Crawley, UK). The VMAT beam configuration was varied (overview and naming see Table 2). For each patient, the clinically accepted treatment plan was evaluated by a physician and served as reference (V2_C15). Different arc types were compared: one single arc (V1_C15), two single arcs (2V1_C15) or one double arc (V2_C15). Two single arcs can be directly created by the user before optimization (2V1_C15) while both arcs are treated equally during optimization. In contrast, one double arc is generated by the system during optimization after only one arc was initially defined by the user (V2_C15). The influence of arc length (full or partial arcs) was evaluated for 10 cases with unilateral located targets (see Table 1). Start and stop angles of partial arcs were set depending on the treated side: 300° to 179° for left sided targets and 181° to 60° for right sided targets.

The impact of collimator setting was explored for two single arcs and double arcs separately. For double arc technique, the impact of different collimator positions was additionally investigated (V2_C15, V2_C40 and V2_C60). The chosen collimator angles (15°, 40°, 60°) are restricted to the first quadrant. The second quadrant is covered by the opposed gantry position in a course of a full rotation. The third and fourth quadrant would be covered by switching X1 and X2.

For two single arcs, the combination of different collimator positions was examined. The combination of collimator positions 15° with 345° was compared to 15° with 60°.This resulted in the comparisons as listed in Table 2 (right column).

The goals for PTVs and OARs were defined based on the evaluation parameter set (see Table 3). Planning structures were created based on desired target coverage and OAR sparing. For simultaneous integrated boost techniques, the difference of target volumes and planning organs at risk was calculated to avoid overlapping structures. The Auto-Planning goals of the clinically accepted treatment plan served as reference for the other techniques. For comparability, target and OAR goals, dose grid and structure definition remained the same for different beam configurations of one patient case. The dose distribution was calculated with collapsed cone algorithm, dose grid resolution was set to 2 mm and a control point spacing of 4° was used. In the Pinnacle³ Auto-Planning module advanced setting parameters were not changed (tuning balance between PTV and OARs 11%, dose fall-off margin 2,2 cm, hot-spot maximum goal 107%). Once the Auto-Planning run was completed, no further adjustment or re-optimization were performed.

Auto-Planning tries to mimic the decision-making process of an experienced operator. During Auto-Planning, individual optimization objectives, constraints and weights are automatically added and adjusted based on the user defined clinical goals. Structures of hot and cold spots are created to compensate for over and under dosage. The Auto-Planning module adjusts iteratively the objective set to best meet the planning goals. During optimization, a certain combination of cold and warm starts is performed. The resulting objective set represent in detail the desired shape of the dose volume histogram (DVH). Nevertheless, each optimization process for any patient-beam-configuration ends potentially in different sets of objectives, because Auto-Planning identifies weak points of the individual plans and tries to compensate them by appropriate additional objectives. For each plan, the composite objective value (COV) is calculated. The COV describes the weighted sum of quadratic deviations of the objective values and the related points in the dose volume histogram. In other words, the COV expresses how good the requirements were fulfilled (target coverage and OAR sparing). The optimization process minimizes COV.

Plan evaluation was based on two approaches: (i) scoring of evaluation parameters and (ii) indirect comparison of the COV using a new cross comparison method. The clinically accepted treatment plan served as reference (V2_C15) for both evaluation methods. The comparison was done for a set of treatment plans for the same patient only.

For the scoring method, an evaluation parameter set was defined in consensus by the physicians of our institution and is considered as standard evaluation set (Table 3). For each patient, the dose limits were scaled according to the prescription dose (D₉₅) in Table 1. Each plan was scored depending on how many requirements were met in the evaluation parameter set. All requirements were equally weighted. The sum of all scores was calculated for each plan. If a plan fulfilled the requirement, this plan achieved the score (see Table 4). If both plans met the requirement and the deviation was more than 1%, the better plan achieved the score. If the deviation was less than 1%, no trial achieved the score (tie). If both trials failed the requirement, the better plan achieved the score.

The second approach compared indirectly the COVs of two plans. The principle of cross comparison is illustrated in a flowchart in Fig. 1. During the optimization step, each technique A creates its own plan A and its corresponding set of objectives S_A. For the evaluation step, the assessment results are expressed by COV(A, S_A). As illustrated in Fig. 1, plan A (generated using technique A) is subjected to its own objective set S_A, resulting in COV(A, S_A). Similarly, plan B creates its own objective set S_B, resulting in COV(B, S_B). The values COV(A, S_A) and COV(B, S_B) cannot be compared directly because the COVs were calculated based on different objective sets (S_A and S_B). Therefore we propose a cross comparison method; the measures S_A and S_B are mutually applied:

Technique A is allowed to set the conditions S_A for assessment of plan B leading to COV(B, S_A). A symmetric situation is effected when also Technique B is allowed to set its valuation standard S_B resulting in COV(A, S_B). Finally, each of both plans A and B is evaluated with two parameter sets (S_A and S_B). As long as each plan wins only according to its own objective set, no decision can be made on which plan is superior (COV(A, S_A) < COV(B, S_A) and COV(B, S_B) < COV(A, S_B)). In contrast, if plan A wins according to both objective sets S_A and S_B, it can clearly be regarded as the better technique: Plan A is better according to its own measure and it is better even if plan B sets the assessment scale. In other words, plan quality of plan A is superior to plan B, if both values of COV(A, S_A) and COV(A,S_B) are lower than COV(B,S_B) and COV(B,S_A). We denote “plan A is superior to plan B” as proposition “F_A > B”; it is true, if

Re-arrangement of Eq. 1 leads to:

An inversion of the left part of Eq. 2 results inwhich is a reformulation of F_A > B.

The difference of scoring was tested using Wilcoxon matched-pair signed rank test. The level of significance was set at 0,05. Plan A is superior to plan B as soon as one of two evaluation methods shows a significant difference between the techniques.

Additional parameters were assessed to investigate how the beam arrangement affects the plan metrics: monitor units (MU), delivery time (T), Paddick conformity index (CI).

The current work was limited to an evaluation of single and double arc techniques based on patient cases for head and neck cancer. Further improvement of plan quality would be expected for using a larger number of arcs or non coplanar beam arrangements. Multiple arc VMAT showed improved plan quality compared to single arcs at cost of increased delivery times and monitor units [14, 15].

As Teoh et al. already mentioned, direct comparisons between different studies are difficult because of significant differences in target volume definitions, target complexity, dose prescription and treatment schedules [20]. An additionally impact will have the treatment planning system and the implemented algorithms for VMAT optimization and segmentation.