| 张桐,刘嘉城,王美娇,等.基于高斯过程的立体定向放射手术计划六维摆位误差鲁棒性评估工具开发与验证[J].中华放射医学与防护杂志,2026,46(3):273-279.Zhang Tong,Liu Jiacheng,Wang Meijiao,et al.Development and validation of a gaussian process-based robustness evaluation tool for 6D setup error in stereotactic radiosurgery treatment planning[J].Chin J Radiol Med Prot,2026,46(3):273-279 |
| 基于高斯过程的立体定向放射手术计划六维摆位误差鲁棒性评估工具开发与验证 |
| Development and validation of a gaussian process-based robustness evaluation tool for 6D setup error in stereotactic radiosurgery treatment planning |
| 投稿时间：2024-12-18 |
| DOI：10.3760/cma.j.cn112271-20241218-00482 |
| 中文关键词: 鲁棒性评估 不确定性量化 立体定向放疗 高斯过程 |
| 英文关键词:Robustness evaluation Uncertainty Stereotactic radiosurgery Gaussian process |
| 基金项目:国家重大研发计划项目(2019YFF01014405);北京大学肿瘤医院科学研究基金项目(ZY202410) |
|
| 摘要点击次数: 3336 |
| 全文下载次数: 67 |
| 中文摘要: |
| 目的 旨在开发针对光子放疗计划的自动化鲁棒性评估工具,为摆位精度要求极高的放射治疗技术(如单中心多靶点立体定向放疗)提供不确定性分析与量化鲁棒性评估结果。方法 本研究采用以高斯过程(GP)为替代模型评估摆位不确定性对剂量指标的影响与计划鲁棒性。通过生成误差场景、计算剂量和收集感兴趣区剂量体积直方图(DVH)指标3个步骤生成数据训练GP模型,最后评估未知场景DVH指标及其对应概率。考虑摆位误差维度包含平移与旋转误差。为评估模型预测精确性,选取3例不同外扩边界(0、1和2 mm)的HyperArc SRS计划进行了鲁棒性分析,并在训练集以外样本上比对TPS计算与GP模型预测的关键DVH指标。结果 GP模型对大部分纳入研究DVH指标的预测较为准确。右 眼平均剂量平均R2为0.97,脑干的最大剂量平均R2为0.96,GTV1的D98%平均R2为0.89。结论 本研究开发的鲁棒性评估工具基于可编程交互接口实现了自动化放疗计划鲁棒性评估,对于摆位误差敏感的治疗技术意义重大。 |
| 英文摘要: |
| Objective To develop an automated robustness evaluation tool for photon radiotherapy plans. The tool is designed to provide uncertainty analysis and quantitative robustness assessment for radiotherapy techniques that require high positioning accuracy, such as single-isocenter multi-target stereotactic radiosurgery. Methods This study employed a Gaussian Process (GP) model as a surrogate to evaluate the impact of setup uncertainties on dose metrics and to assess plan robustness. The data generation process involved three steps: generating error scenarios, computing the corresponding dose distributions, and collecting dose-volume histogram (DVH) metrics to train the GP model. The trained model was then used to predict DVH metrics and their associated probabilities in untested scenarios. The considered setup errors included both translational and rotational dimensions. To validate the predictive accuracy of the model, robustness analyses were conducted on three HyperArc SRS plans with varying margins (0, 1 and 2 mm). Key DVH metrics predicted by the GP model were compared against those calculated by the treatment planning system (TPS) for out-of-sample error scenarios. Results The GP model demonstrated accurate predictions for most DVH metrics investigated. Specifically, the average R2 values were 0.97 for the mean dose to the right eye, 0.96 for the maximum dose to the brainstem, and 0.89 for the D98% of GTV1. Conclusions A robustness evaluation tool for radiotherapy plans was successfully developed in this study. By utilizing a programmable interactive interface, the tool automates the robustness evaluation process, which holds significant importance for treatment techniques that are highly sensitive to setup errors. |
| HTML 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|