Fine-Scale Singularity Detection in Multi-Dimensional Imaging with Regular, Orientable, Symmetric, Frame Atoms with Small Support

具有规则、可定向、对称、小支撑的框架原子的多维成像中的精细奇异性检测

基本信息

批准号：
1720487
负责人：
Emanuel Papadakis
金额：
$ 25万
依托单位：
University of Houston
依托单位国家：
美国
项目类别：
Standard Grant
财政年份：
2017
资助国家：
美国
起止时间：
2017-07-15 至 2022-06-30
项目状态：
已结题

来源：
https://www.nsf.gov/awardsearch/showAward?AWD_ID=1720487&HistoricalAwards=false
关键词：
Fine Scale Singularity Detection Multi

项目摘要

One of life's essential characteristic is movement. Whether it is the spectacular, delicate dance of the unblemished, white swan in Tchaikovsky's Swan Lake, or the early attempts of a toddler to use his or her hands, the neurology of movement is uniquely common for all animals and humans: learning a motor skill, and the necessary muscle coordination. With practice the skill is perfected. Finally, the retainment of this experience-based learning process is the conclusion of this learning process. The ultimate goal of this project is to provide new tools to neuroscientists who study the biological basis of learning at the cell level using live animals. This function is facilitated by a number of anatomical changes in the structure of the cytoplasm of neural cells, such as the formation of lengthy branches known as axons and dendrites, and at a fine scale of dendritic spines and axonal buttons. The latter are less anatomically permanent structures arising on the surface of these cytoplasmic extensions. Dendritic spines and axonal buttons form synapses, which are the communication gateways between neurons. The research team will develop mathematical and computational tools for automatizing the study of spine populations in live neurons, and of their time-evolution during learning. The anticipated outcomes will provide neuroscientists with a number of software tools which will automatize the analysis of synaptic strength and its evolution with learning. These findings will contributed to the better understanding the biological mechanisms of autism and drug addictions.The investigators on this project will develop algorithms for the 3D digital segmentations of dendritic surfaces including spines from 3D images acquired with a certain type of microscope, which uses laser light and works as a scanner by exploiting the natural ability of neurons to fluoresce. They aim to generate accurate binary reconstructions of a dendritic arbor including its spines. The primary challenge in this project is that image acquisition of live neurons has a resolution which provides limited detail of the spines. Often, images contain noise which further complicates the extraction of accurate, binary 3D reconstructions of dendritic surfaces showing spine details. Overcoming this problem is a core goal of the project because spine volume estimation quantifies synaptic strength. These unique challenges lead the investigators to the development of novel mathematical tools for fine scale analysis. They will build ensembles of short in size 3D imaging, 3D-orientation selective, frame-based filters, suitable for sensing curves and surfaces in noisy images. These filters will be designed to respond to local changes of image smoothness. Information obtained from these filters at various scales, will be utilized as input for multilayer, deep-learning inspired neural networks which will determine in an image which voxels belong to spine surfaces. Further algorithmic tools will be developed to track every spine of a dendrite individually over time. The same filtering tools will be used in a different application domain, the generation of illumination neutral images, in real-time. This will help the fast, high throughput removal of the effects of uneven illumination in images inhibiting the detection, by software or the naked eye of contours associated with shapes or textures in a scene. The investigators will develop the mathematical theory of illumination neutralization using concepts from fractal and microlocal analysis. The illumination neutralization algorithm is envisioned to work for real-time video analysis and in conjunction with face verification algorithms with the potential to be used in face recognition, laser microscopy and remote sensing applications.

生命的基本特征之一是运动。无论是柴可夫斯基《天鹅湖》中完美无瑕的白天鹅壮观而精致的舞蹈，还是幼儿早期尝试使用自己的双手，运动的神经学对于所有动物和人类来说都是独一无二的共同点：学习运动技能，以及必要的肌肉协调性。通过练习，技能就会变得完善。最后，这个基于经验的学习过程的保留就是这个学习过程的结论。该项目的最终目标是为神经科学家提供新工具，他们利用活体动物研究细胞水平学习的生物学基础。神经细胞细胞质结构的许多解剖学变化促进了这种功能，例如轴突和树突的长分支的形成，以及树突棘和轴突纽扣的精细尺度。后者是在这些细胞质延伸的表面上产生的解剖学上不太持久的结构。树突棘和轴突按钮形成突触，是神经元之间的通信网关。研究小组将开发数学和计算工具，用于自动研究活神经元中的棘群及其在学习过程中的时间演化。预期的结果将为神经科学家提供许多软件工具，这些工具将自动分析突触强度及其随学习的演变。这些发现将有助于更好地了解自闭症和药物成瘾的生物学机制。该项目的研究人员将开发对树突表面进行 3D 数字分割的算法，其中包括使用某种类型的显微镜获取的 3D 图像中的树突棘，该显微镜使用激光并通过利用神经元发出荧光的自然能力来充当扫描仪。他们的目标是生成树突乔木（包括其脊柱）的精确二元重建。该项目的主要挑战是活体神经元图像采集的分辨率只能提供有限的脊柱细节。通常，图像包含噪声，这使得提取显示脊柱细节的树突表面的精确二元 3D 重建变得更加复杂。克服这个问题是该项目的核心目标，因为脊柱体积估计可以量化突触强度。这些独特的挑战促使研究人员开发用于精细尺度分析的新颖数学工具。他们将构建短尺寸 3D 成像、3D 方向选择性、基于帧的滤波器的集合，适用于感测噪声图像中的曲线和表面。这些滤波器将被设计为响应图像平滑度的局部变化。从这些不同尺度的滤波器获得的信息将被用作多层、深度学习启发的神经网络的输入，该神经网络将确定图像中哪些体素属于脊柱表面。将开发进一步的算法工具来随着时间的推移单独跟踪树突的每个脊柱。相同的过滤工具将用于不同的应用领域，实时生成照明中性图像。这将有助于通过软件或肉眼快速、高吞吐量地消除图像中不均匀照明的影响，从而抑制与场景中的形状或纹理相关的轮廓的检测。研究人员将利用分形和微局域分析的概念开发照明中和的数学理论。照明中和算法预计可用于实时视频分析，并与人脸验证算法结合使用，有可能用于人脸识别、激光显微镜和遥感应用。

项目成果

期刊论文数量（15）

专著数量（0）

科研奖励数量（0）

会议论文数量（0）

专利数量（0）

Stable recovery of planar regions with algebraic boundaries in Bernstein form

DOI：
10.1007/s10444-021-09843-0
发表时间：
2021-02
期刊：
Advances in Computational Mathematics
影响因子：
1.7
作者：
C. Conti;M. Cotronei;D. Labate;Wilfredo Molina
通讯作者：
C. Conti;M. Cotronei;D. Labate;Wilfredo Molina

Blind Image Inpainting with Sparse Directional Filter Dictionaries for Lightweight CNNs

DOI：
10.1007/s10851-022-01119-6
发表时间：
2022-05
期刊：
Journal of Mathematical Imaging and Vision
影响因子：
2
作者：
Jenny Schmalfuss;Erik Scheurer;Hengyuan Zhao;Nikolaos Karantzas;Andrés Bruhn;D. Labate
通讯作者：
Jenny Schmalfuss;Erik Scheurer;Hengyuan Zhao;Nikolaos Karantzas;Andrés Bruhn;D. Labate

Virtual Multimodal Automated Object Detection with Deep Neural Networks

使用深度神经网络的虚拟多模态自动物体检测