Impact of deformable image registration accuracy on thoracic images with different regularization weight parameter settings
Impact of deformable image registration accuracy on thoracic images with different regularization weight parameter settings
复制标题
DOI：
10.1016/j.ejmp.2017.09.122
复制
发表时间：
2017-10-01
影响因子：
3.4
通讯作者：
Nagata, Yasushi
中科院分区：
文献类型：
--
作者：
Miura, Hideharu;Ozawa, Shuichi;Nagata, Yasushi
目的：我们评估了变形图像配准（deformable image registration，简称deformable image registration，简称deformable image registration，简称deformable image registration）的准确性，胸部图像在不同的正则化weights下使用市售的deformable image registration software.Methods：胸部4维（4D）CT图像的10例患者。这些患者的数据集由DIR实验室（www.dir-lab.com）提供，包括手动识别的300个解剖标志的坐标列表。使用RayStation（RaySearch Laboratories，斯德哥尔摩，瑞典）的ANATomically CONstrained Deformation Algorithm（ANACONDA）在不同正则化权重（4、40、400-默认设置，1500、4000、10，000、15，000、20，000、30，000和40，000）下对吸气峰到呼气峰图像进行变形。使用脚本更改正则化权重。通过计算每个标志点处由RISK软件计算的位移与由标志计算的位移之间的差异来确定配准误差（RE）。我们测量了每个正则化权重设置的计算时间。结果：高正则化权重导致比低正则化权重观察到的更小的RE。RE在达到平台之前随着正则化权重的增加而迅速减小。正则化权重为400与正则化权重为4、40、4000或40，000之间没有发现显着差异（P值> 0.05）。平均时间的范围为8.4-12.2 s。结论：我们得出结论，ANACONDA的默认设置相对于胸部区域的正则化权重是稳定的。
Purpose: We assessed the deformable image registration (DIR) accuracy of thoracic images under different regularization weights using commercially available DIR software.Methods: The thoracic 4-dimensional (4D) CT images of 10 patients were used. The datasets for these patients were provided by DIR-lab (www.dir-lab.com) and included a coordinate list of 300 anatomic landmarks that had been manually identified. The ANAtomically CONstrained Deformation Algorithm (ANACONDA) of RayStation (RaySearch Laboratories, Stockholm, Sweden) was used to deform the peak-inhale to peak-exhale images under different regularization weights (4, 40, 400-default setting, 1500, 4000, 10,000, 15,000, 20,000, 30,000, and 40,000). The regularization weights were changed using a script. The registration error (RE) was determined by calculating the difference at each landmark point between the displacement calculated by the DIR software and that calculated by the landmark. We measured the computation time for each regularization weight setting.Results: High regularization weights resulted in a smaller RE than that observed with lower regularization weights. The RE decreases rapidly with increase in regularization weight before reaching a plateau. No significant difference was found between a regularization weight of 400 and regularization weights of 4, 40, 4000 or 40,000 (P value> 0.05). The range of the average time was 8.4-12.2 s.Conclusions: We concluded that the default setting for ANACONDA is stable with respect to regularization weight in the thoracic region.