Next-Generation Calcium Imaging Analysis Methods

项目摘要 钙成像方法使我们能够用单细胞记录多个神经元的同时活动 解决方案；因此，这些方法是大脑倡议和神经科学的关键使能工具 更广泛地说。这些实验产生了巨大的2D或3D视频数据集-在某些情况下具有数据率 以TB/小时为单位--而对这一“大数据”的分析目前是 这一领域的科学进步。该项目开发了强大的新分析方法来消除这种情况 瓶颈，打开了可以用这些新工具攻击的新的科学问题和应用程序。 正在开发的方法同时识别成像神经元的位置，在空间上解析 重叠的神经元形状，并提供每个神经元活动的去噪估计，最小 手动参数调整。新的方法在数量和质量上都比最先进的方法更好 在模拟数据和各种实际数据应用中，导致有用信号的恢复 比其他任何方式都要多得多的神经元。同时，这些方法在计算上是 可扩展和模块化，支持健康的用户和开发社区。最后，这些方法是 可扩展：它们建立在定义良好的概率建模和凸优化原则基础上， 实现了一系列扩展，以解决重要的新科学问题。 该项目的具体目标包括一些关键的子项目，重点放在：第一，扩大这些方法 在计算上尽可能高效地处理非常大的数据集，以实现实时的闭环 实验；第二，加强获得统计最优解的方法，以便提取 尽可能多地从数据中获取信息，并具有尽可能高的时空分辨率，从而实现 新的综合计算成像方法的发展。同时，这个项目将发展 对这些方法的扩展以处理不同的数据类型：空间模糊数据，或通过某些 更复杂的线性成像变换(例如，来自光场相机)；成像数据，其中我们可以 通过利用同时记录的刺激或行为信息来约束和改进我们的结果； 最后，与高时间分辨率多电极电性数据同时记录的成像数据， 以结合这两种数据类型的优点。 拟议的分析工具将在神经科学界广泛使用，并将有很强的 对理解神经科学数据的基本方法的影响；此外，该项目将提供 实验范例，并推动未来的数据收集。

Project Summary Calcium imaging methods allow us to record the simultaneous activity of many neurons with single-cell resolution; these methods are therefore a critical enabling tool for the BRAIN initiative and in neuroscience more broadly. These experiments produce enormous 2D or 3D video datasets – in some cases with data rates measured in terabytes/hour - and the analysis of this “big data” currently represents a major bottleneck on scientific progress in this field. This project develops powerful new analysis methods for eliminating this bottleneck, opening up new scientific questions and applications that can be attacked with these new tools. The methods under development simultaneously identify the locations of the imaged neurons, resolve spatially overlapping neuronal shapes, and provide denoised estimates of the activity of each neuron, with minimal manual parameter tuning. The new methods quantitatively and qualitatively improve upon the state of the art in both simulated data and in a wide variety of real data applications, leading to the recovery of useful signals from many more neurons than otherwise possible. At the same time, the methods are computationally scalable and modular, enabling a healthy user and development community. Finally, the methods are extensible: they are founded on well-defined probabilistic modeling and convex optimization principles, enabling a range of extensions to address important new scientific problems. Specific aims of the project include a number of critical subprojects focused on: first, scaling up these methods to handle very large data sets, as computationally efficiently as possible, to enable closed-loop, real-time experiments; and second, strengthening the methods to obtain statistically optimal solutions, in order to extract as much information from the data as possible, with the highest possible spatiotemporal resolution, enabling the development of novel integrated computational imaging methods. In parallel, this project will develop extensions of these methods to handle different data types: spatially blurred data, or data formed via some more complicated linear imaging transformation (e.g., from light-field cameras); imaging data in which we can constrain and improve our results by exploiting simultaneously-recorded stimulus or behavioral information; and finally, imaging data recorded simultaneously with high-temporal-resolution multielectrode electrical data, in order to combine the strengths of these two data types. The proposed analytical tools will be widely used in the neuroscience community, and will have a strong influence on fundamental approaches to understanding neuroscience data; furthermore, the project will inform experimental paradigms and drive future data collection.