Radiation Oncology, Sir Charles Gairdner Hospital, Nedlands, WA, 6009, Australia.
Australian e-Health Research Centre, CSIRO, Royal Brisbane and Women's Hospital, Herston, QLD, 4029, Australia.
Med Phys. 2019 May;46(5):2243-2250. doi: 10.1002/mp.13494. Epub 2019 Apr 22.
To demonstrate selection of a small representative subset of images from a pool of images comprising a potential atlas (PA) pelvic CT set to be used for autosegmentation of a separate target image set. The aim is to balance the need for the atlas set to represent anatomical diversity with the need to minimize resources required to create a high quality atlas set (such as multiobserver delineation), while retaining access to additional information available for the PA image set.
Preprocessing was performed for image standardization, followed by image registration. Clustering was used to select the subset that provided the best coverage of a target dataset as measured by postregistration image intensity similarities. Tests for clustering robustness were performed including repeated clustering runs using different starting seeds and clustering repeatedly using 90% of the target dataset chosen randomly. Comparisons of coverage of a target set (comprising 711 pelvic CT images) were made for atlas sets of five images (chosen from a PA set of 39 pelvic CT and MR images) (a) at random (averaged over 50 random atlas selections), (b) based solely on image similarities within the PA set (representing prospective atlas development), (c) based on similarities within the PA set and between the PA and target dataset (representing retrospective atlas development). Comparisons were also made to coverage provided by the entire PA set of 39 images.
Exemplar selection was highly robust with exemplar selection results being unaffected by choice of starting seed with very occasional change to one of the exemplar choices when the target set was reduced. Coverage of the target set, as measured by best normalized cross-correlation similarity of target images to any exemplar image, provided by five well-selected atlas images (mean = 0.6497) was more similar to coverage provided by the entire PA set (mean = 0.6658) than randomly chosen atlas subsets (mean = 0.5977). This was true both of the mean values and the shape of the distributions. Retrospective selection of atlases (mean = 0.6497) provided a very small improvement over prospective atlas selection (mean = 0.6431). All differences were significant (P < 1.0E-10).
Selection of a small representative image set from one dataset can be utilized to develop an atlas set for either retrospective or prospective autosegmentation of a different target dataset. The coverage provided by such a judiciously selected subset has the potential to facilitate propagation of numerous retrospectively defined structures, utilizing additional information available with multimodal imaging in the atlas set, without the need to create large atlas image sets.
从包含潜在图谱（PA）盆腔 CT 集的图像池中选择一小部分具有代表性的图像子集，用于自动分割另一个目标图像集。目的是平衡图谱集代表解剖多样性的需求与创建高质量图谱集所需的资源（例如多观察者勾画）之间的平衡，同时保留对 PA 图像集的其他可用信息的访问。
进行图像标准化预处理，然后进行图像配准。聚类用于选择子集，该子集通过配准后图像强度相似性来提供对目标数据集的最佳覆盖。进行了聚类稳健性测试，包括使用不同的起始种子重复聚类运行以及使用随机选择的 90%的目标数据集重复聚类。比较了五个图像（从 39 个盆腔 CT 和 MR 图像的 PA 集中选择）的图谱集对目标集（包含 711 个盆腔 CT 图像）的覆盖情况：（a）随机（在 50 次随机图谱选择中平均），（b）仅基于 PA 集中的图像相似性（代表前瞻性图谱开发），（c）基于 PA 集中的相似性和 PA 与目标数据集之间的相似性（代表回顾性图谱开发）。还比较了整个 39 个图像的 PA 集提供的覆盖范围。
示例选择具有高度的稳健性，示例选择结果不受起始种子选择的影响，只有在目标集减少时，才偶尔会对示例选择之一进行更改。通过目标图像与任何示例图像的最佳归一化互相关相似度来衡量目标集的覆盖范围，由五个精选图谱图像提供（平均值=0.6497）与整个 PA 集（平均值=0.6658）提供的覆盖范围更相似，而不是随机选择的图谱子集（平均值=0.5977）。这对于平均值和分布形状都是如此。回顾性选择图谱（平均值=0.6497）与前瞻性选择图谱（平均值=0.6431）相比仅略有改善。所有差异均具有统计学意义（P<1.0E-10）。
从一个数据集选择一个小的代表性图像集可以用于为不同的目标数据集的回顾性或前瞻性自动分割开发图谱集。这样精心选择的子集提供的覆盖范围有可能利用多模态成像图谱集中的其他可用信息来促进众多回顾性定义的结构的传播，而无需创建大型图谱图像集。
