Strategies to optimize stereotactic radiosurgery plans for brain tumors with volumetric‐modulated arc therapy

Abstract Purpose Prescription practice in SRS plans for brain tumors differs significantly for different modalities. In this retrospective study, the strategies to optimize SRS plans for brain tumors with volumetric arc therapy (VMAT) were presented. Methods Fifty clinically treated cases were stratified by the maximum target size into two groups (≥ 2 cm in 25 cases and <2 cm but ≥1 cm in 25 cases), which were optimized using traditional LINAC MLC‐based approaches with the average prescription isodose line (P‐IDL) of (91.4 ± 0.6)%. Four to five plans have been created for each case with variation of the P‐IDL from 65% to 90%. The optimization strategies to select an optimal P‐IDL, to use tuning structures within the target and beyond as well as to use NTO (normal tissue objectives), were applied to all new plans. Results The optimal P‐IDL was found to be around 75%. After applying the new optimization strategies with an average P‐IDL of 74.8%, the mean modified gradient index (mGI) and V12 were reduced by 25% and 35%, respectively for both groups. The Paddick conformity index (PCI) was averagely improved by 8%. The average homogeneity index (HI) and focal index (FI) were increased by 22% and 50%, respectively. The mGI was inversely proportional to the PTV volumes. The shape of the dose distribution in target was also changed from concave to convex. The comparison of PCI with historical data from other institutes and modalities shows that the plans in this study have the best conformity near the target. Conclusions With the new optimization strategies for VMAT SRS plan of brain tumor more conformal plans in both high and intermediate dose region (~50% of the PD) were created, in which the dose in the core of the target was notably increased while V12 and mGI were significantly decreased, and PCI was improved. The mGI was inversely proportional to the PTV volumes. The optimal P‐IDL is around 75%. The average PCI is the best in this study compared with the published historical data. These strategies are applicable to treatment planning for multiple brain and liver tumors where sparing the tissue peripheral to the target is critical.

the inverse planning and delivery process for SRS treatment.
The importance of dose fall outside of the tumor (intermediate dose region, i.e. 50% of prescription dose region) was omitted in the three RTOG criteria. The assumption implicitly meant that the smaller CI the more conformal of the plan, even though CI explicitly did not include any information for isodose volumes less than the prescription isodose volume. The definition of CI does not consider how much overlap of the volumes between the tumor and the volume covered by the prescription dose (PD). The assumption was incorrect when inverse optimization planning was implemented for IMRT and VMAT as so many variables were introduced. Clark et al 12  helical tomotherapy. However the lack of overall optimization strategies, it was uncertain if the plans were optimal. We developed a set of systematic optimization strategies including using an optimal P-IDL combining with implementing the tuning ring structures within the target and beyond, and implementing the normal tissue objectives. The new strategies were applied to the retrospective study of VMAT SRS plans for brain tumors including metastases and meningioma. The beam isocenter in the treated plans was positioned automatically by the TPS at the center of the tumor. The collimator angle was determined so that the movement direction of HDMLC (high definition MLC) aligning with the minimum dimension of the target for best conformal plan. 14 The treated plans were created with the intention to have more uniform dose in the target and the conventional P-IDL (averagely (91.7 ± 0.5, ±1.7)% of the maximum dose. Results in the tables and text are reported as mean ± 95% confidence intervals, and ± STD.) was applied. All treated plans were 3 or 4 non-coplanar partial RapicArcs and clinically accepted and delivered.

| METHODS & MATERIALS
New optimization strategies for the retrospective study plans were compared with the old ones in Table 2 while keeping the same geometry of the plan, that is, the same number of arcs and range of arcs. The new optimization strategies have three crucial components 15 : 1. Adding the tuning ring structures in the target intends to make a convex dose distribution; 2. Implementing the NTO (normal tissue objectives) improves the conformity in the high dose region; T A B L E 1 Tumor diagnosis and characteristics are listed for a total 50 cases.

Diagnosis
Number of cases The tuning structures are illustrated in Fig. 1. The initial settings of the objectives for the tuning ring structures present in Table 3.
The NTO settings were listed as follows: • Priority: 150 • Distance from target border: 0.03 • Start dose (%): 95 • End dose (%): 40 • Fall-off: 0.8 To find out the optimal P-IDL multiple plans with variation of P-IDL from 65% to 90% have been created for each case. The variation in the P-IDL with the following five indexes were recorded: Paddick comformity index, 16

| RESULTS
The variation of mGI, PCI, and V12 with the P-IDL from 65% to 90% for all the cases, in which the new optimization strategies were applied, are shown in Figs. 2-4, respectively. It can be seen that the change in mGI, PCI, and V12 with P-IDL is relatively small beyond around 75% of P-IDL ( Fig. 2 and Fig. 3). Decreasing the P-IDL further from 75%, the number of MU would increase more with the P-IDL (Fig. 4) while other parameters would not improve significantly.
Therefore overall the optimal range of the P-IDL was determined to be at around 75%, which is comparable with that for conformal static field or dynamic conformal arcs. 3 The HI and FI were inversely linearly proportional to the P-IDL.
Comparison of the five parameters (mGI, V12, PCI, HI, and FI) for plan quality evaluation between the treated and retrospective study plans is presented in Table 4, in which the fifty cases were  Fig. 5. It can be seen that it was a concave shape in the treated plan and a convex shape in the retrospective study plan. The mGI was inversely proportional to PTV volumes as shown in Fig. 6.

| DISCUSSION
For dynamic conformal arc and conformal static treatment planning the P-IDL is adjusted by changing the margin of the field and the dose distribution in the target is more or less inherently determined  ning for brain tumors is not different with that for single tumor except that collimator angle of the gantry might be crucial. 16 Therefore, the new strategies are applicable to the optimization process of VMAT plan for multiple brain tumors.
The PCI value varies little with the PTV size in this study, that is, 0.888 ± 0.01 although the PTV size varied from 1 to 16 cc. Fig. 7 compares the PCI of this study with the historical data from other institutes and modalities such as LINAC circular and mMLC, 18 Cyberknife, 19 Gamma Knife, and RapidArc. 20 The V12, PCI, and mGI have been reported as key indicators for brain toxicities, 1-3 such as radionecrosis, etc. The smaller the V12/  mGI and the larger the PCI, the more is the brain tissue spared and the less the toxicities. Application of the new strategies has increased the core dose in the targets 50% as indicated in FI. In GammaKnife treated cases, in which there is no critical structures inside the target volume, overall there is no study demonstrating a relationship between the maximum dose in the target and complication rates. 16 It was reported 9 in the results of RTOG 9005 that patients treated on LINAC were 2.84 times more likely to have local progression compared to those treated on a Gamma Knife and the most significant difference of the treatment plans between the two modalities is the P-IDL, which was 50% on a Gamma Knife and 80-90% on a LINAC. However this was not confirmed in the report of RTOG 9508. 17 Further research is needed to see if there is correlation between local progression of brain metastases and the P-IDL.
The factors to affect the P-IDL probably are the penumbra of the MLC profiles, the scattering properties of the media (such as brain or lung, etc.) and the extent of modulation of the field. The

| CONCLUSIONS
With the new optimization strategies for VMAT SRS plan of brain tumor more conformal plans in both high and intermediate dose region (~50% of the PD) were created, in which the dose in the core of the target was notably increased while V12 and mGI were significantly decreased, and PCI was also improved. The optimal P-IDL is around 75%. The average PCI is the best in this study compared with the published historical data. These strategies are applicable to treatment planning for multiple brain and liver tumors where sparing the tissue peripheral to the target is critical.

ACKNOWLEDG MENTS
The author thanks Holly Perlingiero, CMD, and Mark Kalman; CMD for help on the patient treatment planning; and Gary Antonelli for support on the research.

CONFLI CT OF INTEREST
There are no conflict of interests.