Session Item

Treatment planning for a brain tumor can be challenging due to the close proximity of numerous critical structures. Our previous study demonstrated a clear benefit of using knowledge-based objectives to drive the plan optimization process. As a next step, the current study explores the potential of multicriteria optimization (MCO) to reduce the dose to critical structures even further. 

Retrospective re-optimization was performed by a single planner using the RapidPlan and MCO tools of Eclipse 15.6 (Varian, Palo Alto, US). The study population consisted of 20 patients prescribed 30×2 Gy for glioblastoma multiforme or meningioma. Target volumes varied in location and size (mean PTV 298 cc, range 121-491 cc). Up to 14 brain and optic structures per patient had been contoured (minimum 11), with at least one structure partially overlapping the PTV. The clinical VMAT setup on a TrueBeam linac was retained in replanning.

The DVH estimation model was described previously. For each patient, the line objectives from RapidPlan were selected for MCO trade-off exploration. MCO generates a series of optimal plans based on the entered objectives, enabling the planner to navigate to a preferred solution.

RapidPlan and MCO plans were compared in terms of mean doses to critical structures. A paired samples t-test was performed per structure type. In addition, PTV doses were assessed by means of a homogeneity index HI = (Dmax - Dmin)/Dmean.

A net improvement was consistently achieved with MCO, but the magnitude varied between patients. In four cases it was possible to reduce the Dmean in each of the contoured structures. The remaining sixteen cases showed more of a trade-off with a few structures receiving a higher Dmean. However, this was always outweighed by a majority of structures receiving a lower Dmean.

For the entire study population, MCO reduced the Dmean in 213 out of 267 contoured structures (80%). The remainder received a higher Dmean (12%) or were unchanged within ±1% (8%).

Separating by structure type, the overall Dmean statistics were improved for 11 out of the 14 structures (p<0.05). The largest difference was observed for the right hippocampus (median value -8.2 Gy: see figure), followed by the optic chiasm (-5.5 Gy) and the pituitary (-3.8 Gy). An increase in Dmean resulted only for the left hippocampus (median value +1.2 Gy).

A trade-off was more evident in the PTV dose distributions. MCO generally yielded lower minima and higher maxima (within clinically acceptable limits). The group mean homogeneity index was 0.30±0.08 and 0.35±0.06 before and after replanning, respectively.

Multicriteria optimization offered a considerable extra benefit in treatment planning for brain tumors. Critical structures could be spared to a greater extent than with knowledge-based optimization alone. The main limiting factor is the PTV dose homogeneity. The combination of MCO with knowledge-based planning is therefore recommended for brain tumor treatment planning.