The development of the human brain is a complex, lifelong process during which collective behaviors of neurons exhibit self-organizing dynamics. Metastable states play a crucial role in understanding the complex dynamical mechanisms of the brain, and analyzing them helps to reveal the mechanisms of functional changes in the brain throughout development and aging. Specifically, global metastable state provides a overall perspective on the flexibility of brain reorganization, while the evolution trajectories of transient functional patterns capture detailed changes in brain activity. The leading eigenvector dynamics analysis (LEiDA) method significantly reduces the dimensionality of data and is widely used to capture the temporal trajectory characteristics of transient functional patterns, i.e., metastable brain states. However, LEiDA's linear dimensionality reduction of highdimensional raw brain data may overlook non-linear information and lose some relationships between features. We developed a framework based on Phase Coherence Graph Autoencoder (PCGAE) that employs graph autoencoders (GAE) for non-linear dimensionality reduction of phase coherence matrices. This approach clusters to identify more distinct metastable brain states and is applied to the analysis of resting-state functional magnetic resonance imaging (rs-fMRI) data across the human lifespan. This paper investigates age-related differences and continuity changes from different perspectives: metastable state indicators and state trajectory indicators (occurrence probability, lifetime, and state transition metrics). Global metastable state shows a linear decline with age, while both linear and quadratic effects of age-related changes are observed in detailed state metastable and state trajectory indicators. Finally, the proposed feature extraction scheme demonstrates good classification performance for categorizing brain age groups. These findings can help us understand the self-organizing reorganization characteristics associated with aging and their complex dynamic changes, providing new insights into brain development throughout the entire lifespan.

功能脑网络中不同的模板定义导致网络规模差异极大,进一步影响所构建网络的结构及其拓扑属性。但是,在机器学习方法中网络规模差异是如何影响特征选择策略及分类准确率并不清楚。研究中采用5种不同节点规模的模板进行脑网络构建,在此基础上选择脑网络的三个局部特征用SVM方法构建分类器进行抑郁症患者的识别。结果表明,节点规模较大的模板的分类准确率较高;同时,在不同节点规模下传统的P值的特征选择方法均是可行的,但其阈值设置过于严格。

由于静息态功能磁共振成像（Resting Functional Magnetic Resonance Imaging,R-f MRI）技术的发展和成熟,现如今已被广泛应用于预测抑郁症等精神疾病中。超网络因能充分反应各个脑区之间的互动而被广泛应用。现有的超网络构建多是基于稀疏表示方法,传统的基于LASSO的稀疏表示方法缺乏解释脑区之间的组效应的能力,因而提出基于分组的超网络构建方法,但仍存在局限性:一方面它们多是基于自动分组,忽视了人脑中复杂的先验组结构;另一方面它们只能进行单级选择而忽略了人脑中复杂的层次关系;另外,由于惩罚函数对系数的过强压缩,导致模型中目标变量回归系数的有偏估计,使得噪声变量在压缩的同时,目标变量也进行了一定程度的压缩。因此,本文考虑到这些问题,并在组效应的基础上,提出三种基于预定义分组的无偏稀疏模型的超网络构建方法以改进原有方法:复合最小最大凹惩罚（Minimax Concave Penalty,MCP）方法、组MCP方法以及组平滑剪裁绝对值偏差（Smoothly Clipped Absolute Deviation,SCAD）方法。第一种方法既能在组间进行变量选择,又能选择组内重要变量,称为双级选择;后两种方法则不具有双级选择能力,只能在组级上进行变量选择。但复合MCP和组MCP方法均是基于最小最大凹惩罚的无偏稀疏模型,而组SCAD是基于平滑剪裁绝对值偏差方法,并将SCAD惩罚应用到组的层面上,从而实现组级选择。实验结果表明,三种方法的超网络在结构上具有差异。组MCP方法和组SCAD方法构建的超网络结构相似,但二者与基于复合MCP方法构建的超网络结构差异较大。分类结果表明复合MCP方法得到较另两种方法好的分类表现。因此在基于预定义分组的超网络构建方法中,复合MCP方法具有较好的解释组效应的能力。本文提出的方法所构建的脑功能超网络可以更好地表达抑郁症患者与正常对照的结构差异,具有重要的理论意义和临床价值。本文的主要创新工作有以下几点:第一,提出基于预定义分组的脑功能超网络构建方法。自动组效应模型只是对具有强相关性的变量实现组选择,没有考虑到人脑中天然形成的组结构。预定义分组模型可对任意变量进行组选择,同时也考虑到人脑中具有相关性的变量。复合MCP、组MCP以及组SCAD方法均通过人为定义分组数目,使用聚类算法将具有相关性的变量划分到特定的组里,重复多次,可对任意变量进行组选择。第二,提出基于无偏稀疏模型的脑功能连接建模方法。复合MCP、组MCP和组SCAD方法均是基于无偏稀疏模型进行超网络构建的。LASSO模型是传统的构建脑网络的方法,但该方法是有偏估计,主要体现在求解稀疏线性模型过程中,对目标变量回归系数的估计上。而LASSO产生有偏估计的原因在于其对噪声变量和目标变量的回归系数的压缩程度是相同的。于是针对LASSO的有偏估计这一缺点,引入了无偏稀疏模型。相比于LASSO方法,三种超网络构建方法所使用的无偏稀疏模型对目标变量回归系数的压缩程度表现出逐渐减小,甚至对目标变量回归系数的压缩彻底消除也是有可能的。第三,提出基于变量的双级选择方法。复合MCP方法在组结构的基础上,引入了双级选择,指既能在组级层面上进行变量选择,又能在组内选择重要的变量,因而变量选择更为严格,但最能反映人脑中复杂的层次结构。结果表明,双级选择的复合MCP方法具有最好的分类表现。本文提出的三种脑功能超网络构建方法,可以有效解决针对LASSO以及以LASSO为基础的各种稀疏表示模型的有偏估计的问题,更好地表达了人脑结构中复杂的层次关系。实验结果表明,基于预定义分组的无偏稀疏模型脑功能超网络构建方法,具有较好的分类表现和分类权重。本文工作受到了国家自然科学基金项目（61472270,61672374,61741212,61876124,61976150）、山西省重点研发计划项目（201803D31043）、山西省科技厅应用基础研究项目青年面上项目（201801D121135,201803D31043）、山西省教育厅高等学校科技创新研究项目（2016139）以及教育部赛尔网络下一代互联网技术创新项目（NGII20170712）的支持。本文的重点是无偏稀疏模型的超网络构建方法对抑郁症患者的分类研究,具有重要的理论意义和临床价值。

Objective: The unclear propagation pathway of seizures is one of the main reasons for failure of surgical treatment, and the propagation process involves the directional brain networks. However, few network analysis techniques have successfully traced specific seizure propagation pathways. This study proposed a stepwise transfer entropy (STE) approach to describe the propagation of effective connections in brain networks. Methods: The proposed STE technique is applied to stereoelectroencephalography (SEEG) data collected from patients with epilepsy, which can identify characteristic regions connected to specific seed brain regions at different link-step levels. Importantly, the underlying TE-based network construction method can accurately obtain electrode-to-electrode connections, and the stepwise approach can capture the interactions between electrodes at the connection level. Finally, simulation and clinical data were used to evaluate the STE approach according to the similarity, confusion matrix, accuracy and recall. Results: We used three datasets: a simulation dataset, a clinical dataset, and a public dataset. We compared the results of different network construction methods with multiple datasets, and the STE approach successfully captured node changes in epilepsy patients, effectively identified early and late propagation nodes, and determined the propagation pathway. Moreover, the STE approach is superior to other methods in all evaluation indices, achieving 97.9% accuracy and 0.93 similarity. Significance: Compared with the existing methods, the STE approach has significant advantages in accurately tracking propagation pathways. Moreover, the STE approach is simple and quickly performs calculations, making it an easy-to-use promising method for determining propagation pathways in clinical settings. © 2023 The Author(s)

BACKGROUND: Resting-state functional magnetic resonance imaging (rs-fMRI) technology and the complex network theory can be used to elucidate the underlying mechanisms of brain diseases. The successful application of functional brain hypernetworks provides new perspectives for the diagnosis and evaluation of clinical brain diseases; however, many studies have not assessed the attribute information of hyperedges and could not retain the high-order topology of hypergraphs. In addition, the study of multi-scale and multi-layered organizational properties of the human brain can provide richer and more accurate data features for classification models of depression.; PURPOSE: This work aims to establish a more accurate classification framework for the diagnosis of major depressive disorder (MDD) using the high-order line graph algorithm. And accuracy, sensitivity, specificity, precision, F1 score are used to validate its classification performance.; METHODS: Based on rs-fMRI data from 38 MDD subjects and 28 controls, we constructed a human brain hypernetwork and introduced a line graph model, followed by the construction of a high-order line graph model. The topological properties under each order line graph were calculated to measure the classification performance of the model. Finally, intergroup features that showed significant differences under each order line graph model were fused, and a support vector machine classifier was constructed using multi-kernel learning. The Kolmogorov-Smirnov nonparametric permutation test was used as the feature selection method and the classification performance was measured with the leave-one-out cross-validation method.; RESULTS: The high-order line graph achieved a better classification performance compared with other traditional hypernetworks (accuracy=92.42%, sensitivity=92.86%, specificity=92.11%, precision=89.66%, F1=91.23%). Furthermore, the brain regions found in the present study have been previously shown to be associated with the pathology of depression.; CONCLUSIONS: This work validated the classification model based on the high-order line graph, which can better express the topological features of the hypernetwork by comprehensively considering the hyperedge information under different connection strengths, thereby significantly improving the classification accuracy of MDD. Therefore, this method has potential for use in the clinical diagnosis of MDD. © 2024 American Association of Physicists in Medicine.

Alzheimer's disease (AD) is a neurodegenerative disease, and mild cognitive impairment (MCI) is considered a transitional stage between healthy aging and dementia. Early detection of MCI can help slow down the progression of AD. At present, there are few studies exploring the characteristics of abnormal dynamic brain activity in AD. This article uses a method called Leading Eigenvector Dynamics Analysis (LEiDA) to study resting-state functional magnetic resonance imaging (rs-fMRI) data of AD, MCI, and cognitively normal (CN) participants. By identifying repetitive states of phase coherence, inter group differences in brain dynamic activity indicators are examined. And the neurobehavioral scales were used to assess the relationship between abnormal dynamic activities and cognitive function. The results showed that in the indicators of occurrence probability and lifetime, the globally synchronized state of the patient group decreased. The activity state of the limbic regions significantly detected the difference between AD and the other two groups. Compared to CN, AD and MCI have varying degrees of increase in default and visual regions activity states. In addition, in the analysis related to the cognitive scales, it was found that individuals with poorer cognitive abilities were less active in the globally synchronized state, and more active in limbic regions activity state and visual regions activity state. Taken together, these findings reveal abnormal dynamic activity of resting-state networks in patients with AD and MCI, provide new insights into the dynamic analysis of brain networks, and contribute to a deeper understanding of abnormal spatial dynamic patterns in AD patients.

为了解决图像压缩感知重建研究领域中通过有效的图像先验信息重构与原图相似性高且保留细节消除伪影的高质量图像的问题,针对不足采样的K空间数据,在经典的CNN算法CBDNet算法的基础上,通过融合深度学习先验信息及传统图像恢复各自优势的方法,研究了基于深度神经网络去噪先验和BM3D块压缩感知算法的混合式重构算法.该算法采用交互式方法训练多尺度残差网络抑制噪声水平,借优化选择的方式将深度学习与传统块匹配多尺度结合以提取图像不同尺度的特征数据从而实现抑制伪影、快速重建高质量MRI.实结果表明深度学习结合BM3D在MR图像重构领域能够有效降低伪影保留细节信息,加强重构效果.与此同时,通过采用GPU的加速运算,算法的计算复杂度较使用单一算法并未增加很多.可见基于卷积盲降噪的混合式核磁共振成像效果更佳.

在阿尔兹海默症分类问题中,超图神经网络可以从被试间的超图关系中提取特征,在表示学习复杂图结构方面具有很好的优势,但大多数模型都直接或间接地将超图所表示的被试间的高阶复杂关系分解,转化为简单的二元关系进行特征学习,没有有效利用超边的高阶信息,因此提出了基于线-超图神经网络（line-hypergraph neural network, L-HGNN）的阿尔兹海默症分类模型,该模型利用稀疏线性回归表征被试间多元相关性,借助超图和线图的转换在神经网络模型中实现节点的高阶邻域信息传递和超边整体结构特征学习,同时,结合注意力机制生成更具区分性的节点嵌入,进而用于阿尔兹海默症的辅助诊断.在ADNI数据上与常用的两种方法比较,实验结果表明,该方法能有效提高分类准确率,在阿尔兹海默症早期诊断上具有重要的应用价值.

利用脑网络模板对脑网络进行构建是当前脑网络疾病领域的一个研究重点,该方法为脑神经疾病的诊断提供了一种更有效的技术。研究发现,不同模板定义下的网络节点规模的差异对网络的结构及其拓扑属性会产生极大影响。然而,传统单一模板构建的网络忽略了节点的规模,也缺乏不同节点规模对拓扑属性可信度影响的探究。同时,对于机器学习方法中网络节点规模的差异是如何影响网络结构、分类准确率的仍在研究中。为解决这些问题,本文在前人的基础上,针对节点规模对网络拓扑属性、分类特征表现、选择策略及分类准确率的影响进行深入研究。以下为本文的主要工作:第一,定义多节点模板并构建网络。本文使用纽约大学公开的正常被试的数据集和山西医科大学抑郁症的数据集分别进行脑区划分,得到五个不同节点数量的脑网络。第二,探究不同节点数量对拓扑属性的影响。对纽约大学公开的正常被试的数据集构建的不同节点规模的脑网络的局部属性进行提取,然后对比拓扑属性的差异,最后对脑网络的功能连接和网络拓扑属性的可靠性进行分析。第三,探究节点规模对分类特征的选择及性能的影响。首先提取抑郁症数据集构建的不同节点规模的脑网络的局部属性,然后通过K-S检验法在局部属性之间进行统计分析,经过判断,选取判别性特征（显著性差异的属性）作为分类特征,采用SVM对抑郁症与正常人进行分类。第四,使用最大相关最小冗余法对特征间的冗余性进行分析,并判断不同节点规模的脑区间的距离对冗余度的影响与传统的统计显著性的特征选择方法在不同节点规模的可行性,最后检验实验中高维特征是否存在过拟合问题。综合考量发现,随着空间节点数量的增加,网络结构的可靠性在逐渐提高,分类效果也在不断改善。结果表明,不同节点规模得到的特征有效性相当,即节点数量多的模板虽然不能提供更有效的特征,但是,却可以提供更多的有效特征,这将导致分类准确率的提升。同时,节点数量多的模板由于脑区间距离更接近,特征间的冗余程度也在增强。文中还发现传统的统计显著性的特征选择方法在不同节点规模下均可行,但分析结果表明,传统的0.05的阈值设置过于严格。本研究得到国家自然科学基金（61672374,61741212,61876124,61873178）,山西省自然科学基金（201601D021073,201801D121135）,山西省高校科技创新项目（2016139）的研究资助。山西省重点研发（R＆D）项目（201803D31043）。本文的研究重点在于多节点模板定义及网络构建,在此基础上进一步探讨了网络节点规模对网络拓扑属性以及分类特征的选择及性能的影响。本文为今后将脑网络拓扑属性应用到机器学习方法时,为构建网络选择合适的模板提供了一定的参考价值。

大脑是一个复杂且精密的器官,尽管人类对大脑的结构以及功能有了一定的认知,但鉴于大脑以及与其相关联器官运作的复杂程度,人类对大脑的探索还在继续。有研究表明,脑疾病的出现与大脑某些区域无法正常工作有关。近年来,核磁共振技术的发展为脑疾病的预防与诊断提供了技术支撑,图论的引入使得脑网络成为了研究的热点。研究者们将脑网络与人工智能领域中应用较广泛的计算机技术如:机器学习、深度学习等结合起来,对脑网络进行频繁子图挖掘、特征提取与选择、分类等工作,进而获得与疾病有关的生物学标志。传统的脑网络研究主要集中在确定图上,该方法忽略了脑网络间的不确定性信息,而且在构建确定图时需要考虑稀疏度选择的问题,但是目前稀疏度的选择并没有黄金标准。因此,有研究者将不确定图应用到大脑的建模中,构建不确定脑网络,它能够很好地体现脑区间的不确定性信息。通过对比已有的不确定脑网络和确定脑网络的研究发现,从分类准确率的角度无法说明使用哪一种模型对大脑建模是有效的。而且,目前对脑网络模型的选择并没有黄金标准。另外,在传统的不确定脑网络研究中,均值、方差、极差等统计指标被用于特征提取。但是,均值存在分类准确率低、泛化性能差的问题;方差反映的是子图模式间的平均差异,其结果是对样本值取平方,无法直接反映子图间的个体差异;极差由计算公式可知它反映的是子图间的最大差异,受最大值最小值影响较大,无法直接衡量子图间的差异。因此,为了解决上述的两个问题,本文的主要工作为:（1）提出了基于独立成分分析的不确定脑网络构建方法考虑到脑网络所具有的不确定性特征,本文将不确定图应用到脑网络建模中。同时,不确定图对应的节点由组独立成分分析方法来确定。该方法采用盲源信号分离方法,加入了组元素,将其应用在群体水平上,其获得的结果具备统计学意义。与传统方法中使用脑图谱对脑部进行划分的方法相比,该方法无先验模板的要求,避免了因为采用不同模板而对实验结果产生的影响,能够更好地反映不同成分间所蕴涵的空间关系。（2）提出了基于判别性子图的不确定脑网络与确定脑网络的对比方法考虑到脑网络的不确定性特征,传统方法中使用的确定图对大脑进行建模忽略了大脑的不确定性特征。但是通过对已有实验中不确定脑网络与确定脑网络的分类结果进行对比,从最终的准确率方面无法说明两种模型的优劣。且目前没有相应的研究表明使用哪一种模型进行脑网络建模的效果更好,这对于以后的研究中两种模型的使用情况没有一个有效的参考。因此,以判别性子图为特征,本文提出了不确定图子图模式稀疏度化和确定图子图模式权值化的双向对比方法,对不确定脑网络模型和确定脑网络模型进行脑疾病的分类性能对比,从分类准确率的角度判断选用哪一种模型对大脑进行建模是合适的且能够有效地对脑疾病进行预测。（3）提出基于相对极差的特征提取方法通过分析已有统计指标均值、方差、极差中存在的问题,本文提出了相对极差作为新的统计指标进行特征提取,它是均值和极差的结合,既考虑到子图模式间的最大差异,又考虑到子图模式间的组间差异,可以有效克服原有方法中统计指标所存在的受最大最小值影响较大以及子图差异无法直接衡量的问题,并结合相应的判别分数函数进行特征选择,最终构建特征矩阵进行分类。结果表明:稀疏度在（0.05-0.25）范围内,不确定脑网络的分类性能优于确定脑网络,而稀疏度在（0.25-0.40）范围内,确定脑网络分类性能优于不确定脑网络。本结论为以后脑疾病诊断中脑网络的模型选择提供了参考。另外,本文提出的相对极差作为新的特征提取方法,其分类性能显著高于已有统计指标均值、方差、极差。且该指标在不同的特征选择方法下表现出较好的分类性能,具有很强的泛化性。本文重点研究了不确定脑网络模型和确定脑网络模型的对比方法,并提出了新的特征提取方法,旨在找到最适合脑网络建模的模型以及最具有判别力的特征,为脑疾病的预防和诊断提供方法。