Näsmark Torbjörn, Andersson Jonas
Department of Radiation Sciences, Radiation Physics, Umeå University, SE-901 85 UMEÅ, Sweden.
Phys Imaging Radiat Oncol. 2023 Sep 20;28:100493. doi: 10.1016/j.phro.2023.100493. eCollection 2023 Oct.
In proton therapy, a 3.5% margin is often used to account for proton range uncertainties, of which computed tomography (CT) image noise is assumed to contribute 0.5%. This work evaluates the noise-sensitivity of three dual-energy computed tomography (DECT)-based methods for mapping proton stopping power relative to water (SPR): Näsmark & Andersson (N&A), Landry-Saito (L-S), and the commercial application DirectSPR.
DECT image data of a CIRS-062M phantom was acquired with CT scanners from two different vendors. Acquisitions were repeated 30 times to account for intra- and inter-scan variations. SPR maps were generated with the three DECT-based methods and range simulated in a commercial treatment planning system.
Noise in input data was amplified in L-S SPR maps, kept level with DirectSPR, while N&A compressed noise overall but displayed sensitivity to the choice of input data, potentially leading to increased noise levels. In our simulations, only N&A improved upon the assumed 0.5% noise contribution to range uncertainty on one scanner. On the other scanner, uncertainties exceeded 0.5% for all three methods. Mitigation of this issue was demonstrated by using a method employing virtual mono-energetic images as input. Increasing imaging radiation dose, as expected, alleviates the problem, while applying noise reduction only helped to a lesser extent.
While range uncertainty due to noise is small compared to other contributions, it becomes more important as we move towards smaller treatment margins and the noise-sensitivity of SPR mapping methods should be carefully estimated and considered before clinical implementation.
在质子治疗中，通常采用3.5%的边界来考虑质子射程的不确定性，其中计算机断层扫描（CT）图像噪声被认为占0.5%。本研究评估了三种基于双能计算机断层扫描（DECT）的用于绘制相对于水的质子阻止本领（SPR）图谱的方法的噪声敏感性：Näsmark & Andersson（N&A）法、Landry-Saito（L-S）法以及商业应用DirectSPR法。
使用来自两家不同厂商的CT扫描仪采集CIRS-062M体模的DECT图像数据。重复采集30次以考虑扫描内和扫描间的变化。使用这三种基于DECT的方法生成SPR图谱，并在商业治疗计划系统中模拟射程。
输入数据中的噪声在L-S SPR图谱中被放大，在DirectSPR图谱中保持不变，而N&A法总体上压缩了噪声，但对输入数据的选择显示出敏感性，这可能导致噪声水平增加。在我们的模拟中，只有N&A法在一台扫描仪上改善了假设的0.5%的噪声对射程不确定性的贡献。在另一台扫描仪上，所有三种方法的不确定性都超过了0.5%。通过使用一种将虚拟单能图像作为输入的方法证明了该问题的缓解。正如预期的那样，增加成像辐射剂量可缓解该问题，而仅应用降噪的帮助较小。
虽然与其他因素相比，噪声引起的射程不确定性较小，但随着我们朝着更小的治疗边界发展，它变得更加重要，并且在临床实施之前应仔细估计和考虑SPR图谱绘制方法的噪声敏感性。
