Biophysical modeling of the functional MRI signal through parametric variations in neuronal activation and blood vessel anatomy using realistic synthetic microvascular networks

非侵入性地测量人类大脑活动的最广泛的工具是功能性磁共振 功能磁共振成像（fMRI），通常使用血氧水平跟踪血流和氧合的变化， 相关（粗体）信号。虽然BOLD是神经放电的间接测量，但它已被证明是一种 然而，脑血管解剖学和生理学的细节是已知的， 影响包括BOLD在内的所有fMRI信号。最近，在动物中的侵入性光学成像研究表明， 与神经元活动同时发生的血流调节的变化远比 以前认为，表明功能磁共振成像可以是一个忠实的代表神经元活动在精细的空间和 时间尺度最近的生物物理模拟进一步证明了微血管网络， 血管对神经活动的反应，可以影响人类的fMRI信号，这表明建模可以 有助于改善功能磁共振成像的解释。我们建议通过一系列生物物理模拟来扩展这项工作， 我们将参数化地改变血管解剖、神经元活动和血管对神经元的反应， 活动，然后模拟产生的BOLD反应，以表征这些对功能磁共振成像的影响。我们假设 血管解剖结构和神经元活动模式的细节都将对 功能磁共振成像信号和我们的建模框架可以预测这些影响-这可以改善推理 功能磁共振成像的神经活动。这种方法现在才有可能，因为有了大规模的 显微镜数据，我们的高效计算框架，我们的新血管合成算法。 在这项工作中，我们将扩展我们新的血流和氧气运输框架，以模拟血管动力学 神经元活动的反应，然后结合MR物理产生相应的BOLD信号。我们 建模平台提供了独特的功能：合成逼真的大规模血管网络， 可控的几何形状、密度和拓扑结构;以及比以往更大的血管系统的鲁棒模拟 企图。这将允许以足够的规模进行准确、高效的计算，以生成有意义的BOLD 与人类功能磁共振成像数据相关的反应。我们将测试血液动力学的其他方面 反应可以提供神经元活动的更忠实的表示。最后，我们将测试我们的模型预测 用一个简单的、高分辨率的人类功能磁共振成像实验来对比经验数据。这项工作有四个目标。在Aim中 1我们比较了四个候选的“场景”描述血管神经活动的反应。在目标2中，我们测试 BOLD通过合成大规模血管网络对血管解剖结构的依赖性。在目标3中，我们测试 通过模拟系统变化的时空，神经元活动模式对BOLD的依赖性 神经元活动的模式在Aim 4中，我们通过高分辨率的人类fMRI测试模型预测 实验测量BOLD响应参数变化的神经元活动模式。这场 工作将是表征功能磁共振成像信号依赖的因素，不能在人体内测量。

The most widespread tool for measuring brain activity noninvasively in humans is functional magnetic resonance imaging (fMRI), which typically tracks changes in blood flow and oxygenation using the blood-oxygenation-level- dependent (BOLD) signal. Although BOLD is an indirect measure of neural firing, it has been shown to be a faithful measure of brain activation, yet the details of brain vascular anatomy and physiology are known to influence all fMRI signals including BOLD. Recently, invasive optical imaging studies in animals demonstrated that the changes in blood flow regulation occurring alongside neuronal activity are far more precise than previously believed, indicating fMRI can be a faithful representation of neuronal activity at fine spatial and temporal scales. Recent biophysical simulations have further demonstrated how the microvascular network, and the vascular response to neural activity, can influence fMRI signals in humans, suggesting that modeling can help improve fMRI interpretation. We propose to extend this work through a series of biophysical simulations in which we will parametrically vary vascular anatomy, neuronal activity, and the vascular response to neuronal activity then simulate the resulting BOLD responses to characterize these influences on fMRI. We hypothesize that the specifics of the vascular anatomy and neuronal activity patterns will both have measurable effects on the fMRI signal and that our modeling framework can predict these influences—which can improve inferences of neural activity from fMRI. This approach is only now possible due to the availability of sufficiently-large-scale microscopy data, our highly efficient computational framework, and our novel vascular synthesis algorithm. For this work we will extend our new blood flow and oxygen transport framework to simulate vasomotive responses to neuronal activity, then incorporate MR physics to generate the corresponding BOLD signals. Our modeling platform provides unique capabilities: synthesis of realistic, large-scale vascular networks with fully controllable geometry, density, and topology; and robust simulations of vascular systems far larger than ever attempted. This will allow for accurate, efficient calculations at a sufficient scale to generate meaningful BOLD responses that can be related to human fMRI data. We will test whether other aspects of the hemodynamic response may provide more faithful representations of neuronal activity. Finally, we will test our model predictions against empirical data with a simple, high-resolution human fMRI experiment. This work spans four Aims. In Aim 1 we compare four candidate “scenarios” describing the vascular response to neural activity. In Aim 2 we test the dependence BOLD on vascular anatomy by synthesizing large-scale vascular networks. In Aim 3 we test dependence of patterns of neuronal activity on BOLD by simulating systematically varying spatiotemporal patterns of neuronal activity. In Aim 4 we test model predictions through a high-resolution human fMRI experiment measuring BOLD responses to parametrically varied neuronal activity patterns. The outcome of this work will be a characterization of fMRI signal dependence on factors that cannot be measured in humans in vivo.