Room: Room 209
Purpose: Single isocenter for multiple targets (SIMT) radiosurgery treatment planning is complex and time consuming. The plan quality is highly dependent on plannerâ€™s experience and technique. This study investigated knowledge-based SIMT planning using volumetric modulated arc therapy (VMAT).
Methods: Treatment plans from 50 patients who had undergone radiosurgery at Duke University Medical Center were selected to train the SIMT optimization model using RapidPlan (Eclipse V13.6, Varian Medical Systems). The 50 training cases were either one fraction (12 â€“ 22 Gy/fx) or five fractions (5.5 Gy/fx) with 1 â€“ 24 targets of volume ranging from <0.1 to 66.3 cmÂ³. The SIMT model was then evaluated and validated with 12 independent cases of 1-17 targets. VMAT plans were copied and re-optimized using knowledge-based SIMT model without user intervention. Dosimetric parameters including planning target volume (PTV) coverage, dose to organs at risk (OARs), conformity index (CI), maximum dose (Dmax), and monitor units (MU) from the model-generated plans were compared to those of clinical plans.
Results: The model-generated plans only went through one round of optimization, which substantially reduced planning time compared to human iterative planning. In general, all 12 model-generated plans are clinically acceptable. The PTV coverage and CI are equivalent to those of clinical plans. Dmax for brainstem, optical chiasm, optical nerves, eyes, and lenses were lower or equivalent to those of clinical plans. Total MUs were significantly reduced (up to 36%) while global Dmax was higher for all 12 model-generated plans compared to the clinical plans.
Conclusion: Though the current model training algorithm does not handle very small structures well, the plans generated with SIMT model were dosimetrically equivalent to the manually generated clinical plans. Knowledge-based planning significantly reduces the optimization time while improving the plan quality and standardization.