Improving the Dose Distribution and Delivery Efficiency in IMRT Inverse Planning by Iteratively Reweighted L1-Norm
H Kim1*, R Li1, L Xing1, (1) Stanford University, Stanford, CA,TU-G-BRB-6 Tuesday 4:30:00 PM - 6:00:00 PM Room: Ballroom B
Purpose: Compressed sensing with L1-norm based total-variation (TV) regularization has been successfully applied in IMRT inverse planning by generating piecewise constant fluence maps. L1-norm was chosen primarily due to its computational efficiency; for sparse signal recovery, an ideal approach is to use L0-norm (the number of non-zero elements) or Lp-norm (p<1). It is, however, difficult to implement in practice as it is a non-convex problem. The goal of this work is to introduce a substitute and yet effective method, which can further reduce the fluence-map complexity and enhance delivery efficiency in IMRT inverse planning, while not damaging the conformal dose distribution.
Methods: We propose to use an algorithm based on iteratively reweighted L1-norm. The algorithm consists of solving a sequence of weighted L1-norm minimization problems where the weights used for the next iteration are computed from the value of the current solution. In this work, a TV solver, called TFOCS, was used for each iteration. The proposed algorithm was evaluated on a prostate case for IMRT inverse planning. The treatment plans were evaluated by the conformation number (CN) and the modulation index (MI).
Results: The iteratively reweighted L1-norm outperforms the L1-norm based TV method for fluence-map optimization in both conformal dose distribution and delivery efficiency. Significantly, the proposed method by reweighting the TV form enables higher or similar conformal dose distribution with lower field complexity. At 60 segments, the complexity of the fluence-map (MI) was reduced by 20%, while producing higher CN. This improvement will be more effective as the dose conformity gets stronger.
Conclusions: The proposed method with the iteratively reweighted L1-norm achieves more conformal dose distribution and improves the delivery efficiency by reducing the fluence-map complexity compared with the conventional TV minimization for IMRT inverse planning.