Université de Lyon, CREATIS, CNRS UMR5220, Inserm U1206, INSA-Lyon, Université Lyon 1, Centre Léon Bérard, Lyon, France.
Med Phys. 2017 Sep;44(9):4548-4558. doi: 10.1002/mp.12448. Epub 2017 Aug 18.
Dual-energy computed tomography (DECT) has been presented as a valid alternative to single-energy CT to reduce the uncertainty of the conversion of patient CT numbers to proton stopping power ratio (SPR) of tissues relative to water. The aim of this work was to optimize DECT acquisition protocols from simulations of X-ray images for the treatment planning of proton therapy using a projection-based dual-energy decomposition algorithm.
We have investigated the effect of various voltages and tin filtration combinations on the SPR map accuracy and precision, and the influence of the dose allocation between the low-energy (LE) and the high-energy (HE) acquisitions. For all spectra combinations, virtual CT projections of the Gammex phantom were simulated with a realistic energy-integrating detector response model. Two situations were simulated: an ideal case without noise (infinite dose) and a realistic situation with Poisson noise corresponding to a 20 mGy total central dose. To determine the optimal dose balance, the proportion of LE-dose with respect to the total dose was varied from 10% to 90% while keeping the central dose constant, for four dual-energy spectra. SPR images were derived using a two-step projection-based decomposition approach. The ranges of 70 MeV, 90 MeV, and 100 MeV proton beams onto the adult female (AF) reference computational phantom of the ICRP were analytically determined from the reconstructed SPR maps.
The energy separation between the incident spectra had a strong impact on the SPR precision. Maximizing the incident energy gap reduced image noise. However, the energy gap was not a good metric to evaluate the accuracy of the SPR. In terms of SPR accuracy, a large variability of the optimal spectra was observed when studying each phantom material separately. The SPR accuracy was almost flat in the 30-70% LE-dose range, while the precision showed a minimum slightly shifted in favor of lower LE-dose. Photon noise in the SPR images (20 mGy dose) had lower impact on the proton range accuracy as comparable results were obtained for the noiseless situation (infinite dose). Root-mean-square range errors averaged over all irradiation angles associated to dual-energy imaging were comprised between 0.50 mm and 0.72 mm for the noiseless situation and between 0.51 mm and 0.77 mm for the realistic scenario.
The impact of the dual-energy spectra and the dose allocation between energy levels on the SPR accuracy and precision determined through a projection-based dual-energy algorithm were evaluated to guide the choice of spectra for dual-energy CT for proton therapy. The dose balance between energy levels was not found to be sensitive for the SPR estimation. The optimal pair of dual-energy spectra was material dependent but on a heterogeneous anthropomorphic phantom, there was no significant difference in range accuracy and the choice of spectra could be driven by the precision, i.e., the energy gap.
双能 CT（DECT）已被提出作为单能 CT 的有效替代方法，以降低将患者 CT 数转换为组织相对于水的质子阻止本领比（SPR）的不确定性。本工作旨在使用基于投影的双能分解算法，通过模拟 X 射线图像来优化适用于质子治疗计划的 DECT 采集协议。
我们研究了不同电压和锡过滤组合对 SPR 图准确性和精度的影响，以及在低能（LE）和高能（HE）采集之间分配剂量的影响。对于所有光谱组合，使用具有实际能量积分探测器响应模型的 Gammex 体模对虚拟 CT 投影进行了模拟。模拟了两种情况：没有噪声的理想情况（无限剂量）和对应于 20 mGy 总中央剂量的泊松噪声的实际情况。为了确定最佳剂量平衡，对于四种双能光谱，将 LE 剂量相对于总剂量的比例从 10%变化到 90%，同时保持中央剂量不变。使用两步基于投影的分解方法来得出 SPR 图像。从重建的 SPR 图中分析确定了 70 MeV、90 MeV 和 100 MeV 质子束在 ICRP 成人女性（AF）参考计算体模上的范围。
入射光谱之间的能量分离对 SPR 精度有很大影响。最大程度地增加入射能隙可降低图像噪声。然而，能隙不是评估 SPR 准确性的良好指标。在分别研究每个体模材料的情况下，SPR 准确性的最佳光谱具有很大的可变性。在 LE 剂量范围为 30-70%时，SPR 准确性几乎呈平坦状态，而精度则显示出略微偏向较低 LE 剂量的最小值。在 SPR 图像（20 mGy 剂量）中，光子噪声对质子射程准确性的影响较小，因为在无噪声情况下（无限剂量）获得了可比的结果。与无噪声情况相比，与双能成像相关的所有照射角度的平均均方根射程误差分别为 0.50 毫米和 0.72 毫米，实际情况分别为 0.51 毫米和 0.77 毫米。
通过基于投影的双能算法评估了双能光谱和能级之间剂量分配对 SPR 准确性和精度的影响，以指导质子治疗用双能 CT 的光谱选择。在 SPR 估计中，能级之间的剂量平衡并不敏感。最佳的双能光谱对是材料依赖性的，但在异质的拟人体模上，射程准确性没有明显差异，光谱的选择可以由精度（即能隙）驱动。
