Advanced signal processing and systems identification for investigating the brain's resting-state

提出的研究旨在开发最先进的生理信号和系统分析方法，重点是数据驱动的方法，即它们不依赖于对系统结构的先前假设，并且能够量化动态非线性和/或时变行为。虽然设想的方法贡献具有普遍适用性，但我们将特别考虑定制它们以应用于脑血流动力学和脑静息状态网络数据分析。我们将主要集中在采用基展开概念的建模方法上，因为这种方法显著减少了所需的自由参数数量（特别是对于非线性系统），并且它们已被证明在生物工程相关应用中是成功的。具体来说，它们已被广泛用于生物系统建模，包括在快速发展的功能神经成像领域的研究。提出的研究计划的主要目的是：(i)通过结合机器学习和更传统的系统识别原理（如贝叶斯估计、神经网络架构和递归估计方案），优化非线性和/或非平稳系统建模的基展开技术。从长远来看，我们的目标是提供一个通用框架，将非线性系统识别与统计/机器学习技术联系起来。（ii）应用时变技术研究非平稳性对脑血流动力学和人脑静息状态网络分析的影响。为此，我们将使用已有的多种模式（功能磁共振成像、同时进行的脑电图-功能磁共振成像、脑磁图）的数据，以及在蒙特利尔神经学研究所脑成像中心收集的长时间数据，包括同时进行的脑电图-功能磁共振成像和同时进行的脑电图-脑磁图。（iii）利用上述多模态数据研究静息状态网络在多个尺度上的特征，利用每个模态的优势（fMRI的空间分辨率与EEG和MEG的时间分辨率）；（iv）通过使用生物物理和数据驱动模型以及在MR扫描仪中收集相关生理信号（压力、脉搏波速度、脑血流速度），将经颅多普勒超声提供的血流动力学和自动调节措施与功能神经成像（特别是功能磁共振成像）相结合。预计这将增强对脑血流动力学两个方面的解释（多普勒超声产生的系统和功能磁共振成像产生的区域），并通过进一步将其分解为自主和认知因素，以更清晰的方式阐明静息状态网络活动的功能作用。拟议研究的影响预计是显著的，因为它将提供更好的理解脑血流动力学和静息状态脑功能网络，两者在临床研究中表现出显著的生物标志物潜力。

The proposed research aims to develop state-of-the art methods for the analysis of physiological signals and systems, focusing on methods that are data-driven, i.e. they do not rely on prior assumptions about system structure, and are able to quantify dynamic nonlinearities and/or time-varying behavior. While the envisioned methodological contributions are of general applicability, we will specifically consider customizing them for application to cerebral hemodynamics and brain resting-state network data analysis. We will mostly concentrate on modeling approaches that employ the concept of basis expansions, as such approaches significantly reduce the required number of free parameters (especially for nonlinear systems) and that they have proven to be successful in bioengineering-related application. Specifically they have been widely used for modeling biological systems, including those studied in the rapidly developing field of functional neuroimaging. The main aims of the proposed research program are: (i) to optimize basis expansions techniques for modeling nonlinear and/or nonstationary systems by combining principles from machine learning and more traditional systems identification, such as Bayesian estimation, neural network architectures and recursive estimation schemes respectively. In the longer term we aim to provide a general framework that will link nonlinear systems identification with statistical/ machine learning techniques. (ii) to apply time-varying techniques to investigate the effect of nonstationarities on cerebral hemodynamics and human brain resting-state network analyses. To this end we will use already existing data from multiple modalities (fMRI, simultaneous EEG-fMRI, MEG) as well as collect long duration data at the Brain Imaging Center of the Montreal Neurological Institute, including simultaneous EEG-fMRI as well as simultaneous EEG-MEG. (iii) to investigate the characteristics of resting-state networks over multiple scales by using the aforementioned multimodal data, which exploit the advantages of each modality (space resolution of fMRI with time resolution of EEG and MEG), as well as simulations from biophysical (forward models) (iv) integrate measures of hemodynamics and autoregulation provided by transcranial Doppler ultrasound with functional neuroimaging (particularly fMRI) by using biophysical and data-driven models as well as by collecting relevant physiological signals (pressure, pulse wave velocity, cerebral blood flow velocity) in the MR scanner. This is anticipated to enhance the interpretation of both aspects of cerebral hemodynamics (systemic as yielded by Doppler ultrasound and regional as yielded by fMRI) and to elucidate the functional role of resting-state network activity in a clearer manner, by further disentangling it into autonomic and cognitive factors. The impact of the proposed research is expected to be significant as it will provide better understanding of cerebral hemodynamics and resting-state brain functional networks, with both exhibiting significant potential as biomarkers in clinical studies.