Implementation and dissemination of cloud-based retrospective hemodynamic analysis tools to enhance HCP data interpretation

摘要 两年多来，功能磁共振成像数据一直是神经科学研究的支柱 几十年，因为它允许快速、连续、非侵入性地监测神经元功能。然而，一个实质性的 部分fMRI信号来源于低位和心脏的纯生理性脑血流动力学信号 频段。从历史上看，这些只是被认为是使fMRI分析复杂化的噪声源。 我们已经开发了两种新的、回顾性的分析来分离神经和血流动力学部分 大胆的fMRI数据不仅可以模拟和消除低频和心脏系统噪声，而且可以利用 这种“噪声”可以高精度地表征大体积和脉动的血流。这些技术已经被 经过广泛的测试和验证，并在处理来自多个 消息来源。第一种技术，累进时间延迟的回归插值法(“激流”)，分离和 描述了fMRI数据[2]中的低频全局血流动力学信号；由于这是血液传播信号， Riptie可用于测量正常和病理患者的血液到达时间和整个大脑的rCBV 传阅[3-8]。不需要单独的灌注扫描。此外，这产生了体素噪声回归，它 去除混淆信号而不产生由全局信号回归引起的虚假相关性[10]， 大大提高了静息状态和任务fMRI分析检测神经元的特异性，而不是 血流动力学[10，11]。第二种技术是“Happy”，它可以回溯地提取体积图 来自多频段fMRI数据的信号[12]，即使在 没有记录(或记录失败)体积图的受试者，以及标测和/或移除 心脏在大脑中移动时的脉搏波形。这两个工具共同增强和扩展 通过消除相当大比例的以前难以处理的带内噪声，同时提供 新的、完全独立于现有数据的脑血流动力学和自主神经功能窗口。两者都有 已经在开源的“激流”包中发布。 该项目将使这些工具更上一层楼，改进它们的文档、功能和可靠性， 对它们进行优化，以便对多个HCP和ABCD数据集进行回溯分析，并使其可用 通过开发基于云的平台来支持来自各种规模机构的用户(不仅是 资金雄厚、拥有广泛计算基础设施的大型大学)来使用该软件。

Summary Functional Magnetic Resonance Imaging data has been a mainstay of neuroscience research for more than two decades, as it allows rapid, continuous, noninvasive monitoring of neuronal function. However, a substantial portion of the fMRI signal arises from purely physiological cerebral hemodynamic signals in the low and cardiac frequency bands. Historically, these have simply been considered noise sources that complicate fMRI analysis. We have developed two novel, retrospective analyses to separate the neuronal and hemodynamic portions of BOLD fMRI data to not only model and remove low and cardiac frequency systemic noise, but to make use of this “noise” to characterize bulk and pulsatile blood flow with high precision. These techniques have been extensively tested and validated and have shown great flexibility in processing fMRI data from a number of sources. The first technique, Regressor Interpolation at Progressive Time Delays (“RIPTiDe”), isolates and characterizes the low frequency global hemodynamic signal in fMRI data[2]; as this is a bloodborne signal, RIPTiDe can be used to measure blood arrival time and rCBV throughout the brain in normal and pathological circulation[3-8]. without a separate perfusion scan. Moreover, this generates voxelwise noise regressors which remove confound signal without generating the spurious correlations arising from global signal regression[10], substantially increasing the specificity of resting state and task fMRI analyses for detecting neuronal, rather than hemodynamic[10, 11]. The second technique, “happy”, allows retrospective extraction of the plethysmogram signal from multiband fMRI data[12], allowing R-R interval and heart rate variability measurements even in subjects where no plethysmogram was recorded (or failed to record), and mapping and/or removing of the cardiac pulsation waveform as it moves through the brain. These two tools together both enhance and extend existing datasets by removing a substantial proportion of previously intractable in-band noise, while providing new, entirely separate windows into cerebral hemodynamics and autonomic function from existing data. Both have been released in the open source “rapidtide” package. This project will take these tools to the next level, by improving their documentation, capabilities and reliability, optimizing them for retrospective analysis of the multiple HCP and ABCD datasets, and making them available to the widest audience by developing a cloud-based platform to allow users from institutions of all sizes (not only large, well-funded universities with extensive computing infrastructure) to use this software.