The Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology and Emory University, Health Sciences Research Building, Atlanta, GA, United States.
Front Neural Circuits. 2022 Apr 4;16:681544. doi: 10.3389/fncir.2022.681544. eCollection 2022.
Resting-state functional MRI (fMRI) exhibits time-varying patterns of functional connectivity. Several different analysis approaches have been developed for examining these resting-state dynamics including sliding window connectivity (SWC), phase synchrony (PS), co-activation pattern (CAP), and quasi-periodic patterns (QPP). Each of these approaches can be used to generate patterns of activity or inter-areal coordination which vary across time. The individual frames can then be clustered to produce temporal groupings commonly referred to as "brain states." Several recent publications have investigated brain state alterations in clinical populations, typically using a single method for quantifying frame-wise functional connectivity. This study directly compares the results of k-means clustering in conjunction with three of these resting-state dynamics methods (SWC, CAP, and PS) and quantifies the brain state dynamics across several metrics using high resolution data from the human connectome project. Additionally, these three dynamics methods are compared by examining how the brain state characterizations vary during the repeated sequences of brain states identified by a fourth dynamic analysis method, QPP. The results indicate that the SWC, PS, and CAP methods differ in the clusters and trajectories they produce. A clear illustration of these differences is given by how each one results in a very different clustering profile for the 24s sequences explicitly identified by the QPP algorithm. PS clustering is sensitive to QPPs with the mid-point of most QPP sequences grouped into the same single cluster. CAPs are also highly sensitive to QPPs, separating each phase of the QPP sequences into different sets of clusters. SWC (60s window) is less sensitive to QPPs. While the QPPs are slightly more likely to occur during specific SWC clusters, the SWC clustering does not vary during the 24s QPP sequences, the goal of this work is to improve both the practical and theoretical understanding of different resting-state dynamics methods, thereby enabling investigators to better conceptualize and implement these tools for characterizing functional brain networks.
静息态功能磁共振成像（fMRI）表现出功能连接的时变模式。已经开发了几种不同的分析方法来检查这些静息态动力学，包括滑动窗口连接（SWC）、相位同步（PS）、共激活模式（CAP）和准周期模式（QPP）。这些方法中的每一种都可以用来生成随时间变化的活动或区域间协调模式。然后可以对单个帧进行聚类，以产生通常称为“脑状态”的时间分组。最近的几项出版物研究了临床人群中的脑状态改变，通常使用单一方法来量化逐帧功能连接。本研究直接比较了 k-均值聚类与这三种静息态动力学方法（SWC、CAP 和 PS）的结果，并使用人类连接组计划的高分辨率数据量化了多个指标的脑状态动力学。此外，通过检查第四种动力学分析方法 QPP 识别的脑状态重复序列中脑状态特征的变化，比较了这三种动力学方法。结果表明，SWC、PS 和 CAP 方法在产生的聚类和轨迹上有所不同。这些差异的一个明显说明是，每种方法如何为 QPP 算法明确识别的 24s 序列产生非常不同的聚类分布。PS 聚类对 QPP 敏感，大多数 QPP 序列的中点被分组到同一个聚类中。CAP 也对 QPP 非常敏感，将 QPP 序列的每个相位分为不同的聚类集。SWC（60s 窗口）对 QPP 的敏感性较低。虽然 QPP 更有可能在特定的 SWC 聚类中发生，但 SWC 聚类在 24s QPP 序列期间不会变化，这项工作的目的是提高不同静息态动力学方法的实际和理论理解，从而使研究人员能够更好地构思和实施这些工具来描述功能脑网络。
