Room: Exhibit Hall
Purpose: A fast Born iterative inverse planning approach for VMAT is proposed in this work.
Methods: Within each iteration of the inverse algorithm, two optimization steps are carried. The first step find the photon intensity level per gantry angle by means of least square minimization while assumes fixed MLC positions. The second step seeks for the optimal MLC positions via level-set approach while assuming constant photon intensity levels. To accelerate the iterative algorithm, two photon dose calculation models are involved in the process. The first model is built based on photon pencil beam kernel. The second model store doses that are pre-computed by the pencil beam engine while it only block or keep doses shadowed by a particular beamlet or MLC pixel. The second model is faster and less accurate, those it will be involved mainly through the first iterations. As the iteration evolves, doses will be computed mainly using the pencil beam engine to allow further accurate results. An initialization process is created that generates open leaf MLC for pixels that corresponds to beamlets crossing over PTV while avoiding OARs. Due to the shortcoming in accuracy of the fast dose calculation model, a smarter optimization procedure is implemented. The use of q-norm minimization with "q" closer or equal to unity for the dose misfit term will be involved.
Results: The fast model is at least time 10 times faster if compared to the pencil beam engine. The use of the initialization process helps avoiding to trap in a local minima and accelerates the convergence of the algorithm.
Conclusion: The proposed approach is promising in term of achieving fast VMAT planning with less conformability error to the prescribed doses.
Optimization, Inverse Planning, Dose Volume Histograms
TH- External beam- photons: VMAT dose optimization algorithms