An automated algorithm to detect the trabecular-cortical bone interface in micro-computed tomographic images
An automated algorithm to detect the trabecular-cortical bone interface in micro-computed tomographic images
复制标题
DOI：
10.1007/s00223-007-9063-8
复制
发表时间：
2007-10-01
影响因子：
4.2
通讯作者：
Judex, Stefan
中科院分区：
文献类型：
--
作者：
Lublinsky, Svetlana;Ozcivici, Engin;Judex, Stefan
微计算机断层扫描（microCT）已成为临床前研究中骨形态评价的标准工具。遗憾的是，将骨小梁与皮质骨分开的轮廓线的用户依赖性定义不仅极其耗时，而且还可能代表数据偏差和增加的可变性的来源。在这里，开发了一种自动图像分割技术，并在大范围的骨表型上进行了测试。该算法的主要步骤包括模糊，在不同的阈值分割，和体积成分标记，首先识别骨膜边缘，然后创建一个皮质掩膜，其内边缘定义的小梁-皮质界面。该算法针对（1）8名熟练的microCT操作员进行了测试，这些操作员手动定义了4只成年小鼠股骨远端内的骨小梁，以及（2）单个用户在71只啮齿动物股骨中绘制的轮廓线。在四个股骨中，使用者之间的变异系数为骨体积分数9%，连接密度13%，骨小梁厚度3%。由该算法产生的形态测量数据在8个操作员的平均值的2%以内。在71个股骨中，除骨小梁厚度外，形态测定自动和用户数据之间回归的斜率和截距分别与1和0无显著差异。由于该算法与目前的黄金标准技术的良好匹配，该算法可能会为骨形态的标准化和自动化评价提供一个有价值的工具，而无需耗时的绘制轮廓线的任务。
Micro-computed tomography (microCT) has become a standard tool for the evaluation of bone morphology in preclinical studies. Unfortunately, the user-dependent definition of contour lines that separate trabecular from cortical bone is not only extremely time-consuming but may also represent a source of data bias and increased variability. Here, an automated image segmentation technique was developed and tested over a large range of bone phenotypes. The principal steps of the algorithm involve blurring, segmentation at different thresholds, and volumetric component labeling to first identify the periosteal edge and then create a cortical mask, the inner edge of which defines the trabecular-cortical interface. The algorithm was tested against (1) eight skilled microCT operators who manually defined the trabecular bone within the distal femur of four adult mice as well as (2) contour lines drawn by a single user in femurs from 71 rodents. Across the four femurs, the coefficient of variation between users was 9% for bone volume fraction, 13% for connectivity density, and 3% for trabecular thickness. Morphometric data produced by the algorithm were within 2% of the mean values of the eight operators. Across the 71 femurs, the slope and intercept of the regressions between morphometric automatic and user data were, with the exception of trabecular thickness, not significantly different from 1 and 0, respectively. Because of the excellent match with the current gold-standard technique, this algorithm may present a valuable tool for the standardized and automated evaluation of bone morphology without the time-consuming task of drawing contour lines.