摘要目的:探究一种以剂量体积直方图参数为优化变量，使用多目标优化算法的剂量优化配准算法，以提高锥形线束CT（CBCT）图像验证匹配的准确性。方法:回顾性分析哈尔滨医科大学附属肿瘤医院2022年收治的6例肺癌和5例宫颈癌患者的28套CBCT图像。以骨性配准的结果为剂量配准算法的初始点，在其周围的三维空间内计算各位移点的通量加权灰度均方差，并使用无监督k均值聚类方法筛选候选位移点。使用有限尺寸笔形束算法计算各候选位移点的三维剂量分布，并提取直方图剂量指标作为多目标优化算法的优化变量。使用多目标遗传算法求解Pareto最优解集后，根据预设的目标权重方案求解最优唯一解。结果:采用剂量优化配准后，肺癌病例计划靶区（PTV）的D 90%、D 95%、D 98%、D mean、适形度（CI）指标，较常规配准策略分别平均提高0.23 Gy、0.49 Gy、1.05 Gy、0.15 Gy、0.03，危及器官的剂量差异无统计学意义。宫颈癌病例PTV的D 90%、D 95%、D 98%、D mean、CI指标，较常规配准策略分别平均提高0.72 Gy、1.15 Gy、2.53 Gy、0.24 Gy、0.05，部分危及器官评估指标降低1.06~1.81 Gy。 结论:剂量优化配准方法可有效提高靶区覆盖度并降低危及器官剂量，减少刚性配准算法的残余误差，可作为自适应放疗技术的流程之一。

abstractsObjective:To improve the accuracy of cone-beam CT (CBCT) image registration by using a dose-guided registration algorithm based on multi-objective optimization.Methods:A total of 28 sets of CBCT images of 6 patients with lung cancer and 5 patients with cervical cancer admitted to Harbin Medical University Cancer Hospital in 2022 were retrospectively analyzed. Using the results of bone-based registration as the starting points for dose registration algorithm, the dose fluence weighted mean square errors of each displacement point in surrounding three-dimensional space were calculated, and the candidate displacement points were selected by unsupervised k-means clustering method. The three-dimensional dose distribution of each candidate displacement point was calculated by using the limited size pencil beam algorithm, and the dose histogram indexes were extracted as the optimization variables of the multi-objective optimization algorithm. After the Pareto optimal solution set obtained using multi-objective genetic algorithm, the optimal unique solution was then determined based on the predetermined objective weight scheme.Results:After dose optimization registration, the values of D 90%, D 95%, D 98%, D mean and conformity index (CI) of planning target volume (PTV) of lung cancer patients were increased by 0.23 Gy, 0.49 Gy, 1.05 Gy, 0.15 Gy, 0.03 compared with conventional registration, respectively, and no significant difference was found for the organs at risk (OAR). For cervical cancer cases, the values of D 90%, D 95%, D 98%, D mean and CI of PTV were increased by 0.72 Gy, 1.15 Gy, 2.53 Gy, 0.24 Gy, 0.05 compared with conventional registration, respectively, whereas the evaluation indexes of partial OAR were decreased by 1.06-1.81 Gy. Conclusion:The proposed dose-guided registration algorithm can improve the dose coverage for the target area, decrease the dose for OAR and reduce residual error of rigid registration algorithm, which can be implemented as part of online adaptive radiotherapy.