XuvTools: free, fast and reliable stitching of large 3D datasets
XuvTools: free, fast and reliable stitching of large 3D datasets
复制标题
DOI：
10.1111/j.1365-2818.2008.03094.x
复制
发表时间：
2009-01-01
影响因子：
2
通讯作者：
Burkhardt, H.
中科院分区：
文献类型：
--
作者：
Emmenlauer, M.;Ronneberger, O.;Burkhardt, H.
目前的生物医学研究越来越需要以高分辨率成像大而厚的3D结构。突出的例子是在脑片中远距离追踪细丝，或者在秀丽线虫或斑马鱼等整个动物中定位基因表达或细胞迁移。为了同时获得高分辨率和大视场(FOV)，多个记录(平铺)的组合是选项之一。虽然已经有了快速和可重现地获取多个3D瓦片的硬件解决方案，但通用的软件解决方案还不能快速准确地组装这些瓦片。在本文中，我们提出了一个框架，只要提供瓦片之间的一些小重叠，就可以实现记录在3D空间中任意位置的瓦片的全自动重组。实现了所有瓦片之间的全自动3D关联，从而不需要人工交互或关于瓦片位置的先验知识。我们使用(1)多尺度方法中的仅相位相关来估计粗略位置，(2)对在显著点处提取的小块进行归一化互相关以获得精确匹配，(3)通过奇异值分解找到所有瓷砖的全局最优位置，(4)通过在瓷砖边缘进行漂白校正来实现几乎无缝的拼接。如果数据集包含多个通道，则使用所有通道来获取平铺之间的最佳匹配。为了加速，我们使用启发式方法来剪除不需要的相关性，并通过快速傅立叶变换(FFT)计算所有相关性，从而获得非常好的运行时间性能。我们展示了所提出的框架在从整个斑马鱼胚胎和线虫、小鼠和大鼠脑切片和细毛(毛状体)的广泛的不同数据集上的成功应用。此外，我们将我们的拼接结果与其他商业和免费获得的软件解决方案的结果进行了比较。提出的算法作为开源工具集‘XUVTool’在相应作者的网站http://lmb.informatik.uni-freiburg.de/people/ronneber)，上免费提供，该网站获得了GNU通用公共许可证v2的许可。提供了用于Linux和Microsoft Windows的二进制文件。该工具集是用模板化的C++编写的，因此它可以操作任何位深度的数据集。由于随后使用64位寻址，可以缝合任意大小(即大于4 GB)的堆栈。标准台式计算机的运行时间在几分钟的范围内。还提供了用于高级手动交互和可视化的用户友好界面。
Current biomedical research increasingly requires imaging large and thick 3D structures at high resolution. Prominent examples are the tracking of fine filaments over long distances in brain slices, or the localization of gene expression or cell migration in whole animals like Caenorhabditis elegans or zebrafish. To obtain both high resolution and a large field of view (FOV), a combination of multiple recordings ('tiles') is one of the options. Although hardware solutions exist for fast and reproducible acquisition of multiple 3D tiles, generic software solutions are missing to assemble ('stitch') these tiles quickly and accurately.In this paper, we present a framework that achieves fully automated recombination of tiles recorded at arbitrary positions in 3D space, as long as some small overlap between tiles is provided. A fully automated 3D correlation between all tiles is achieved such that no manual interaction or prior knowledge about tile positions is needed. We use (1) phase-only correlation in a multi-scale approach to estimate the coarse positions, (2) normalized cross-correlation of small patches extracted at salient points to obtain the precise matches, (3) find the globally optimal placement for all tiles by a singular value decomposition and (4) accomplish a nearly seamless stitching by a bleaching correction at the tile borders. If the dataset contains multiple channels, all channels are used to obtain the best matches between tiles. For speedup we employ a heuristic method to prune unneeded correlations, and compute all correlations via the fast Fourier transform (FFT), thereby achieving very good runtime performance.We demonstrate the successful application of the proposed framework to a wide range of different datasets from whole zebrafish embryos and C. elegans, mouse and rat brain slices and fine plant hairs (trichome). Further, we compare our stitching results to those of other commercially and freely available software solutions.The algorithms presented are being made available freely as an open source toolset 'XuvTools' at the corresponding author's website http://lmb.informatik.uni-freiburg.de/people/ronneber), licensed under the GNU General Public License (GPL) v2. Binaries are provided for Linux and Microsoft Windows. The toolset is written in templated C++, such that it can operate on datasets with any bit-depth. Due to the consequent use of 64bit addressing, stacks of arbitrary size (i.e. larger than 4 GB) can be stitched. The runtime on a standard desktop computer is in the range of a few minutes. A user friendly interface for advanced manual interaction and visualization is also available.