School of Physics, Beihang University, Beijing, 102206, People's Republic of China.
National Cancer Center/ National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, 100021, People's Republic of China.
Phys Med Biol. 2023 Jul 21;68(15). doi: 10.1088/1361-6560/ace49b.
. Proton source model commissioning (PSMC) is critical for ensuring accurate dose calculation in pencil beam scanning (PBS) proton therapy using Monte Carlo (MC) simulations. PSMC aims to match the calculated dose to the delivered dose. However, commissioning the 'nominal energy' and 'energy spread' parameters in PSMC can be challenging, as these parameters cannot be directly obtained from solving equations. To efficiently and accurately commission the nominal energy and energy spread in a proton source model, we developed a convolution neural network (CNN) named 'PSMC-Net.'. The PSMC-Net was trained separately for 33 energies (, 70-225 MeV with a step of 5 MeV plus 226.09 MeV). For each, a dataset was generated consisting of 150 source model parameters (15 nominal energies ∈ [,+ 1.5 MeV], ten spreads ∈ [0, 1]) and the corresponding 150 MC integrated depth doses (IDDs). Of these 150 data pairs, 130 were used for training the network, 10 for validation, and 10 for testing.. The source model, built by 33 measured IDDs and 33 PSMC-Nets (cost 0.01 s), was used to compute the MC IDDs. The gamma passing rate (GPRs, 1 mm/1%) between MC and measured IDDs was 99.91 ± 0.12%. However, when no commissioning was made, the corresponding GPR was reduced to 54.11 ± 22.36%, highlighting the tremendous significance of our CNN commissioning method. Furthermore, the MC doses of a spread-out Bragg peak and 20 patient PBS plans were also calculated, and average 3D GPRs (2 mm/2% with a 10% threshold) were 99.89% and 99.96 ± 0.06%, respectively.. We proposed a nova commissioning method of the proton source model using CNNs, which made the PSMC process easy, efficient, and accurate.
. 质子源模型校准（PSMC）对于确保使用蒙特卡罗（MC）模拟进行笔形束扫描（PBS）质子治疗中的精确剂量计算至关重要。PSMC 的目的是使计算剂量与实际剂量相匹配。然而，在 PSMC 中对“标称能量”和“能散”参数进行校准时可能会遇到挑战，因为这些参数无法通过求解方程直接获得。为了高效、准确地对质子源模型中的标称能量和能散进行校准，我们开发了一种名为“PSMC-Net”的卷积神经网络。PSMC-Net 分别针对 33 种能量（70-225 MeV，步长为 5 MeV 加上 226.09 MeV）进行了训练。对于每种能量，生成了一个由 150 个源模型参数（15 个标称能量 ∈[，+1.5 MeV]，10 个能散[0，1]）和相应的 150 个 MC 积分深度剂量（IDD）组成的数据集。在这 150 对数据中，有 130 对用于训练网络，10 对用于验证，10 对用于测试。该源模型由 33 个测量的 IDD 和 33 个 PSMC-Nets（用时 0.01 s）构建，用于计算 MC IDD。MC 和测量的 IDD 之间的伽马通过率（GPR，1mm/1%）为 99.91±0.12%。然而，在未进行校准的情况下，相应的 GPR 降低至 54.11±22.36%，这突出了我们的 CNN 校准方法的巨大意义。此外，还计算了扩展布拉格峰和 20 个患者 PBS 计划的 MC 剂量，平均 3D GPR（2mm/2%，阈值为 10%）分别为 99.89%和 99.96±0.06%。我们提出了一种使用 CNN 对质子源模型进行新的校准方法，使 PSMC 过程变得简单、高效和准确。
