Automated reconstruction of neurites from 3D microscopy image stacks

从 3D 显微镜图像堆栈自动重建神经突

• 批准号: 7586655
• 负责人: ARMEN STEPANYANTS
• 金额: $ 27.3万
• 依托单位: NORTHEASTERN UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2008
• 资助国家: 美国
• 起止时间: 2008-04-01 至 2013-03-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/7586655
• 关键词: Algorithms; Animals; Axon; Brain; Caenorhabditis elegans; Computer software; Dendrites; Dendritic Spines; Development; Disease; Funding; Goals; Image; Label; Length; Manuals; Methods; Microscopy; Morphology; Neuraxis; Neurites; Neurons; Plug-in; Population; Procedures; Process; Public Domains; Pyramidal Cells; Research; Seeds; Specific qualifier value; Structure; Synapses; Syncope; System; Time; Vertebral column; base; flexibility; graphical user interface; hippocampal pyramidal neuron; neural circuit; neuronal cell body; reconstruction; three dimensional structure; tool; two-photon

DESCRIPTION (provided by applicant): Currently, accurate methods of analysis of neuron morphology are based on manual or semi-automated tracing systems. Such tracings can be time consuming and/or are prone to errors in situations where faint or beaded neurites diffusely cover large volumes. This is usually the case with tracing axons of cortical pyramidal neurons, e.g. long range horizontal projections. With this proposal we aim to develop a tool which will automate the reconstruction process of neurites from 3D microscopy stacks of images. The existence of such a tool is critical for advancing neural circuits research. As axons of many neuron types can span the entire brain of an animal (e.g. cortical pyramidal cell axons) or the entire animal itself (e.g. C. elegans), our ultimate goal is to perform reconstructions on a large scale to recover axonal and dendritic arbors of sparsely labeled populations of neurons in their entirety. Our algorithm consists of two main parts. First, a 3D stack of images is segmented into regions based on a local watershed type segmentation procedure. For this, preferred orientations are calculated in each voxel of the thresholded stack of images by applying a bank of steerable 3D Gabor filters. Regions are grown by stepping down in intensity and placing edges between adjacent voxels with dissimilar orientations. Second, created regions are merged into larger structures using global optimization criteria. Here, optimal connecting paths are determined for every pair of regions by maximizing the intensity along the path and, at the same time, keeping the path length to a minimum. Regions are merged depending on the intensity and curvature along their optimally connecting paths. The specific aims of this proposal are as follows. Specific Aim 1: We will develop a graphic user interface (GUI) and optimization based algorithm for the semi-automated tracing of neurites from 3D microscopy stacks of images. The algorithm will be based on the gradient ascent method for finding optimal paths which connect user specified seed points. The GUI will provide the user fast and flexible control over the details of the procedure. We will develop this semi-automated reconstruction tool to function as an autonomous unit, but the GUI and the tracing algorithm are also essential parts of the Specific Aim 2. Specific Aim 2: We will develop a segmentation based algorithm for a fully-automated tracing of neurites from 3D microscopy stacks of images. The algorithm will use watershed type segmentation combined with global optimization based criteria for merging the segmented regions. The fully-automated algorithm will be implemented in the GUI and will utilize methods developed as part of the Specific Aim 1. Specific Aim 3: We will complete the reconstruction process by automatically detecting neuron cell bodies, branching structure, axonal boutons, and dendritic spines. The GUI will provide an opportunity to correct possible errors by connecting and disconnecting branches, removing and adding branches, spines, and boutons. Simple morphometric functions, such as the calculation of length and numbers of boutons and spines, will be implemented as well. Currently, accurate methods of quantitative analysis of neuron morphology and synaptic connectivity are based on manual or semi-automated tracing tools which are time consuming and can be prone to errors. With this proposal we aim to develop a tool that will fully-automate the reconstruction process of neurites from 3D microscopy stacks of images. The existence of such a tool is critical for advancing basic neural circuits research and understanding changes in the central nervous system which underlie its disease state.

描述（由申请人提供）：目前，神经元形态的准确分析方法是基于手动或半自动跟踪系统。这样的描记可能是耗时的和/或在模糊或珠状神经突弥漫地覆盖大体积的情况下容易出错。这通常是皮质锥体神经元的追踪轴突的情况，例如长距离水平投射。有了这个建议，我们的目标是开发一种工具，将自动重建过程中的神经突从3D显微镜堆栈的图像。这种工具的存在对于推进神经回路研究至关重要。由于许多神经元类型的轴突可以跨越动物的整个大脑（例如皮质锥体细胞轴突）或整个动物本身（例如C. elegans），我们的最终目标是进行大规模的重建，以完整地恢复稀疏标记的神经元群体的轴突和树突乔木。我们的算法包括两个主要部分。首先，基于局部分水岭类型分割过程将3D图像堆栈分割成区域。为此，通过应用一组可操纵的3D Gabor滤波器，在图像的阈值化堆叠的每个体素中计算优选取向。通过逐步降低强度并在具有不同取向的相邻体素之间放置边缘来生长区域。其次，使用全局优化标准将创建的区域合并为更大的结构。这里，通过使沿着路径的强度沿着最大化并且同时使路径长度保持为最小来为每对区域确定最佳连接路径。区域根据其最佳连接路径的强度和曲率沿着合并。这项建议的具体目标如下。具体目标1：我们将开发一个图形用户界面（GUI）和基于优化的算法，用于从3D显微镜图像堆栈中半自动跟踪神经突。该算法将基于梯度上升的方法来寻找连接用户指定的种子点的最佳路径。GUI将为用户提供对程序细节的快速灵活控制。我们将开发这种半自动重建工具，使其作为一个自主单元运行，但GUI和跟踪算法也是Specific Aim 2的重要组成部分。具体目标2：我们将开发一种基于分割的算法，用于从3D显微镜图像堆栈中全自动跟踪神经突。该算法将使用分水岭型分割结合全局优化的标准合并分割的区域。全自动算法将在GUI中实现，并将利用作为特定目标1的一部分开发的方法。具体目标3：我们将通过自动检测神经元细胞体、分支结构、轴突终扣和树突棘来完成重建过程。GUI将提供通过连接和断开分支、删除和添加分支、脊和按钮来纠正可能错误的机会。简单的形态测量功能，如计算的长度和数量的扣和脊椎，也将实现。目前，神经元形态和突触连接性的定量分析的准确方法是基于手动或半自动跟踪工具，这是耗时的，并且可能容易出错。有了这个建议，我们的目标是开发一种工具，将完全自动化的重建过程中的神经突从3D显微镜堆栈的图像。这种工具的存在对于推进基本神经回路研究和理解中枢神经系统的变化至关重要，这些变化是疾病状态的基础。