TH‐A‐213AB‐10: Improved Multi‐Criteria Optimization for Intensity Modulated Proton Therapy Using Iterative Resampling of Randomly Placed Pencil‐Beams

Purpose: In treatment planning for spot‐scanned intensity modulated proton therapy (IMPT), a fine‐grid pencil‐beam distribution is used to guarantee high‐quality treatment plans. This may lead to very large optimization problems with excessive planning times, especially for large target volumes. To improve the trade‐off between plan quality and optimization times, we have developed a new pencil‐beam placement method called ‘resampling’. Methods: Resampling is based on repeated multi‐criteria optimizations. In each iteration, a new sample of randomly placed pencil‐beams is optimized together with favorable (high‐weight) pencil‐beams of the previous solution. In previous studies, resampling was successfully applied for Cyber Knife plan optimization. In this study, IMPT resampling plans for four head‐and‐ neck cancer patients were compared with traditional IMPT plans, generated using a regular grid with typical spacing (5×5×4mm³). For resampling, sample sizes of 3000, 5000, 7000, and 9000 pencil‐beams per iteration were tested, storing the (intermediate) plans after each of in total 7 iterations. The same dose prescription and 3‐beam arrangement was used in all plans. Results: For similar optimization times and target coverage, resampling resulted in lower doses to organs‐at‐risk than the traditional approach. Mean doses were reduced by 0.4Gy on average (1.7%, range: 0Gy–0.9Gy) for both parotid glands, by 2.5Gy (7.1%, range 0.5Gy– 3.8Gy) for both submandibular glands and by 3.3Gy (9.6%, range: 2.3Gy–4Gy) for the swallowing muscles. Maximum doses to spinal cord and brain stem were on average reduced by 4.1Gy (25.7%, range: 2.3Gy–5.8Gy) and 3.4Gy (31.4%, range:−0.4Gy–9.5Gy), respectively. For comparable doses to organs‐at‐risk, optimization time was reduced by 38.7% on average (range: 6.7%–54.8%), being proportional to the target volume. Conclusions: Pencil‐beam resampling is an efficient method for IMPT plan optimization, allowing for clinically relevant improvements in plan quality and/or planning time, especially for large problem sizes. It opens possibilities for dealing with large‐scale problems such as beam‐angle optimization.