Performance of Knowledge-Based Radiation Therapy Planning for the Glioblastoma Disease Site.

Abstract

The presence of multiple serial organs at risk (OARs) in close proximity to the tumor makes treatment planning for glioblastoma (GBM) complex and time consuming. The present study aimed to create a knowledge-based (KB) radiation therapy model for GBM patients using RapidPlan. An initial model was trained using 82 glioblastoma patients treated with 60 Gy in 30 fractions. Plans were created using either volumetric modulated arc therapy (VMAT) or intensity modulated radiation therapy (IMRT). To improve the goodness-of-fit of the model, an intermediate model was generated by using the dose-volume histograms (DVHs) of best spared OARs of the initial model. Using the intermediate model and manual refinement, all 82 cases were replanned, resulting in the final model. The final model was validated on an independent set of 45 patients with GBM, astrocytoma, oligodendroglioma, and meningioma. The plans created by the final model exhibited superior planning target volume (PTV) dose metrics compared with manual clinical plans: ΔD_99%=-0.52 ± 0.20 Gy, and ΔD_1%=0.80 ± 0.13 Gy (differences are computed as clinical-model). OAR maximum doses were statistically similar, with improved optic apparatus sparing (ΔD_max=2.78 ± 0.82 Gy). Stated improvements correspond to P<.05. The KB planning time is typically 7 minutes for IMRT and 13 minutes for VMAT, compared with a typical 4 hours for manual planning. The KB approach results in significant improvement in planning efficiency and in superior PTV coverage and better normal tissue sparing irrespective of tumor size and location within the brain.