兰州大学机构库 >信息科学与工程学院
基于脑电图的功能脑网络分析方法及抑郁识别研究
Alternative TitleResearch on EEG-based function brain network analysis and depression recognition
邵学晓
Subtype博士
Thesis Advisor胡斌
2023-05-21
Degree Grantor兰州大学
Place of Conferral兰州
Degree Name工学博士
Degree Discipline计算机应用技术
Keyword脑电图 Electroencephalography 功能脑网络 function brain network 抑郁识别 depression recognition 生物标志物 biomarkers 注意力机制 attention mechanism
Abstract

大脑功能的实现主要依赖于神经元之间的相互作用形成的大规模网络。近年来,基于神经影像数据的功能网络研究逐渐成为了脑科学研究的重点方向之一。抑郁症是一种严重的精神障碍,其主要临床特征表现为个体情感和认知功能的损害。大量神经影像学证据表明抑郁症患者的情绪认知障碍与大脑特定区域间的异常功能连接密切相关。特别是基于脑电图(Electroencephalogram,EEG)功能脑网络的研究发现抑郁症患者存在异常的网络拓扑和连接,这对于探索抑郁症患者的异常神经机制具有重要的意义和作用。然而,针对基于EEG功能脑网络的相关方法及抑郁识别等研究仍然存在一些不足和挑战。主要包括以下几方面:(1)基于EEG功能脑网络构建方法尚无公认的“黄金标准”,难以准确评估和反映大脑活动状态。(2)当前研究者多从单一尺度或者水平进行探索,缺乏从多个尺度和层次获取大脑功能相关信息的研究。(3)大多数的抑郁症临床干预研究缺少客观、有效的生物学指标验证,这使得其病理学的可解释性稍显不足。(4)基于EEG的抑郁识别模型大都忽视了EEG电极通道的区域分布特征,这阻碍了新型计算模型的开发和优化。针对上述的问题和挑战,本研究通过理论驱动和数据驱动两种互补的方法解决上述问题,主要工作内容总结如下:

第一,针对当前基于EEG功能脑网络构建方法尚无“黄金标准”的问题,本研究尝试建立一种可靠、准确的功能网络构建方法,并将其应用于抑郁症患者大脑功能网络的分析。主要包含两个工作:一是对九种传统功能连接方法的理论背景和原理进行了系统性探索和分析,实验结果发现基于相干性的功能网络在抑郁识别研究中几乎总能取得最佳的分类性能。二是提出了一种加权功能超网络方法来表征EEG信号的高阶关系。首先,基于稀疏回归方法构建超网连接;其次,将每个超边节点之间的Person相关系数进行叠加,并将其作为超边的权重值,从而构建包含超边信息的加权超网络;最后,提取网络度量特征进行分类。实验结果显示,超网方法的分类精度优于传统的功能网络方法。本研究尝试从对比传统耦合方法和开发超网络方法两种思路来探索基于EEG功能脑网络的构建方法,这可能为抑郁症患者功能脑网络的理论研究提供了有价值的参考。

第二,针对目前功能脑网络研究只在单一尺度或水平上进行分析的缺点,本研究提出了一种基于改进经验模式分解(Empirical mode decomposition,EMD)的功能网络分解模型,该模型适用于功能脑网络的时频分析。该方法一方面解决了传统EMD方法处理脑电数据引起的模式混叠问题。另一方面,基于不同固有模态函数分别构建功能连接矩阵,实现对抑郁症患者功能脑网络的时频分析。相比于传统的分频段分析,实验结果发现抑郁症患者大脑功能连通性在不同固有模态函数上表现出更加显著的变化,而且基于网络度量的统计结果表明,抑郁组与正常对照组之间存在显著的统计学差异,这些都证实了抑郁症患者存在异常的网络拓扑结构。基于功能脑网络的时频分析可能为抑郁症患者的脑功能连通性分析提供了新的思路。

第三,针对临床干预研究大多缺乏客观的生物学指标来评价其治疗效价的不足,本研究首次从大脑功能连接的角度去分析难治性抑郁障碍患者接受一氧化二氮(也称笑气)治疗的影响。我们首先探索了难治性抑郁障碍患者治疗前后不同时期事件相关电位成分(P1、N1、P2、N2、P3)的变化。实验结果发现笑气治疗前后难治性抑郁障碍患者在P1、N1、P2、N2成分的波幅和潜伏期上具有不同程度的显著差异。其次,探索了笑气处理前后患者功能脑网络的变化,并对其网络度量指标的变化与临床特征进行相关性分析。实验结果发现笑气治疗在一定程度上增强了患者的大脑功能连通性。此外,实验还发现了笑气干预组中聚类系数的增加与抑郁程度的下降显著相关。这些发现为理解笑气如何调节难治性抑郁障碍患者的脑功能提供了神经影像学证据,并为其临床应用提供了参考。

第四,已有功能脑网络的相关研究发现不同脑区上的区域性信息或有助于抑郁症潜在生理特征的挖掘。而现有抑郁识别研究大都忽略了EEG电极通道的区域分布特征。因此,本研究提出了一种基于注意力机制的抑郁识别方法。通过设计分-总架构Transformer模型,来捕获EEG信号中的抑郁特异性信息,从而提高抑郁识别精度。本文通过采用留一交叉验证的方法比较了新模型与现有模型的性能。实验结果表明,同一数据集上与其变体和基线方法相比,新模型可以实现89.7%的最佳分类性能。基于EEG数据的空间分布特征来引导模型内数据流动的思路可能为今后的抑郁识别模型研究提供了新的启示。

综上所述,本文从理论驱动和数据驱动两个方面开展基于EEG的功能脑网络分析方法及抑郁识别研究。首先,探索了多种基于EEG功能网络构建方法,并在不同维度和多尺度水平上分析抑郁症患者功能网络的异常连接模式,从而揭示抑郁症患者的异常神经机制;此外,结合病理学原理来评价抑郁症患者的治疗效果,探究其潜在的神经通路;最后,根据EEG的时空特性来优化抑郁识别建模,从而为抑郁症患者的辅助诊断提供更加可靠的方法和生物标志物。

Other Abstract

The implementation of brain functions mainly relies on the large-scale network formed by the interaction between neurons. In recent years, research on functional brain networks based on neuroimaging data has gradually become a focal point in the field of neuroscience. Depression is a serious mental disorder characterized by impairments of individual emotional and cognitive functions. Numerous neuroimaging studies have provided evidence indicating that emotional and cognitive impairments in depression are associated with abnormal function and structure in certain brain regions. Specifically, research on functional brain networks based on electroencephalogram (EEG) has identified abnormal network topology and connectivity in depressed patients, which is important for understanding and uncovering the underlying neural mechanisms of depression. However, there are still some challenges and limitations in research on EEG-based functional brain networks analysis and depression identification. These challenges mainly include: (1) there is no recognized "gold standard" for constructing functional brain networks based on EEG, which is difficult to accurately evaluate and reflect brain activity; (2) current researchers mainly focus on a single scale or level, making it difficult to simultaneously obtain relevant information on brain functions from multiple scales and levels; (3) most clinical intervention studies on depression lack objective and effective biological indicators for validation, which makes its pathological interpretability slightly insufficient; (4) EEG-based depression identification models have mostly overlook the regional distribution characteristics of EEG electrode channels, which hinders the development and optimization of new computational models. In response to these problems and challenges, this study attempts to solve them through complementary approaches of theory-driven and data-driven methods. The main work is summarized as follows:

First, since there is no "gold standard" for constructing functional brain networks based on EEG, this study attempts to establish a reliable and accurate method for constructing functional brain networks, which are applied to the analysis of functional brain networks in depressed individuals. This mainly includes two tasks: first, through a systematic exploration of nine traditional functional connectivity methods and experiments, it was found that coherence-based functional networks almost always achieved the best classification performance in depression identification. Second, a weighted functional hypernetwork method was proposed to characterize high-order relationships in EEG signals. Firstly, a sparse regression method was used to construct hyper-network connections; Secondly, the Pearson correlation coefficients between each pair of hyper-edge nodes were aggregated and used as the weight of the hyper-edge to better integrate hyper-edge information, thereby constructing a weighted hyper-network that includes hyper-edge information. Finally, network metrics features were extracted for classification. The experimental results showed that the classification accuracy of the hyper-network method is superior to traditional functional network methods. This study attempts to explore EEG-based functional brain network construction methods from the perspective of comparing existing coupling methods and developing hyper-network methods, which may provide valuable reference for the study of functional brain network in patients with depression.

Second, addressing the limitation of current functional brain network research that only focuses on analysis at a single scale or level, this study proposes a functional network decomposition model based on improved Empirical Mode Decomposition (EMD) for time-frequency analysis of functional brain networks. On the one hand, this method solves the problem of pattern aliasing caused by traditional EMD methods in processing EEG data. On the other hand, by constructing brain functional networks with different intrinsic mode functions (IMF), the study achieves time-frequency analysis of the functional brain networks for depression. Compared to traditional frequency band analysis, the experimental results found that the connectivity of brain function in depression patients showed more significant changes in different intrinsic mode functions. More statistically significant results based on network measures suggested that there were significant differences between the depression group and the normal control group. These findings confirm the abnormal network topology in patients with depression. The time-frequency analysis based on functional brain networks may provide new ideas for the analysis of brain functional connectivity in patients with depression.

Third, a limitation of many clinical intervention studies is the lack of objective biological markers to evaluate treatment efficacy. This study is the first to analyze the impact of nitrous oxide (also known as laughing gas) treatment on patients with treatment-resistant depression from the perspective of brain functional connectivity. Firstly, we explore the changes in event-related potential components (P1, N1, P2, N2, P3) at different time points before and after treatment in patients with treatment-resistant depression. The experimental results show significant differences in the amplitude and latency of P1, N1, P2, and N2 components before and after nitrous oxide treatment in patients with treatment-resistant depression. Secondly, we examine changes in functional brain networks before and after nitrous oxide treatment, and correlates changes in network metric indicators with clinical features. The experimental results show that nitrous oxide treatment increases functional connectivity in the brains of patients to a certain extent. In addition, the study found a significant correlation between the increase in clustering coefficient in the nitrous oxide intervention group and the decrease in depression severity. These findings provide neuroimaging evidence for understanding how nitrous oxide regulates the brain function of patients with treatment-resistant depression and provide reference for its clinical application.

Fourth, previous studies on functional brain networks have found that regional information from different brain regions may be used to explore the potential physiological characteristics of depression. However, most existing studies on depression identification have overlooked the regional distribution characteristics of EEG electrode channels. Therefore, this study proposes a depression recognition method based on attention mechanism. By designing a Divide-general Structure Transformer model to capture depression specific information in EEG signals, the accuracy of depression recognition can be improved. The performance of the new model was compared with other models using leave-one-out cross-validation. Experimental results showed that the new model achieved the best classification performance of 89.7% on the same dataset compared with its variants and baseline methods. By using the spatial distribution characteristics of EEG data to guide the flow of data within the model, this approach may provide new insights for future depression recognition model research.

In summary, this study systematically explores the methods for analyzing functional brain networks and depression identifying based on EEG data from both a theoretical and data-driven perspective. Firstly, it various EEG-based functional network construction methods were explored, and abnormal connection patterns of functional networks in depression patients were analyzed at different dimensions and multi-scale levels to reveal the abnormal neural mechanisms of depression patients; In addition, combining pathological principles to evaluate the therapeutic effect of depression patients and explore their potential neural pathways; Finally, it optimizes depression recognition modeling based on the spatiotemporal characteristics of EEG, in order to provide more reliable methods and biomarkers for auxiliary diagnosis of depression patients.

MOST Discipline Catalogue工学 - 计算机科学与技术(可授工学、理学学位) - 计算机应用技术
URL查看原文
Language中文
Other Code262010_120190907271
Document Type学位论文
Identifierhttps://ir.lzu.edu.cn/handle/262010/539187
Collection信息科学与工程学院
Affiliation
兰州大学信息科学与工程学院
Recommended Citation
GB/T 7714
邵学晓. 基于脑电图的功能脑网络分析方法及抑郁识别研究[D]. 兰州. 兰州大学,2023.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Altmetrics Score
Google Scholar
Similar articles in Google Scholar
[邵学晓]'s Articles
Baidu academic
Similar articles in Baidu academic
[邵学晓]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[邵学晓]'s Articles
Terms of Use
No data!
Social Bookmark/Share
No comment.
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.