Evidence of networks in the resting-brain reflecting the spontaneous brain activity is perhaps the most significant discovery to understand intrinsic brain functionality. Moreover, subsequent detection of dynamics in these networks can be milestone in differentiating the normal and disordered brain functions. However, capturing the correct dynamics is a challenging task since no ground truths' are present for comparison of the results. The change points of these networks can be different for different subjects even during normal brain functions. Even for the same subject and session, dynamics can be different at the start and end of the session based on the fatigue level of the subject scanned. Despite the absence of ground truths, studies have analyzed these dynamics using the existing methods and some of them have developed new algorithms too. One of the most commonly used method for this purpose is sliding window correlation. However, the result of the sliding window correlation is dependent on many parameters and without the ground truth there is no way of validating the results. In addition, most of the new algorithms are complicated, computationally expensive, and/or focus on just one aspect on these dynamics. This study applies the algorithms and concepts from signal processing, image processing, video processing, information theory, and machine learning to analyze the results of the sliding window correlation and develops a novel algorithm to detect change points of these networks adaptively. The findings in this study are divided into three parts: 1) Analyzing the extent of variability in well-defined networks of rodents and humans with sliding window correlation applying concepts from information theory and machine learning domains. 2) Analyzing the performance of sliding window correlation using simulated networks as ground truths for best parameters’ selection, and exploring its dependence on multiple frequency components of the correlating signals by processing the signals in time and Fourier domains. 3) Development of a novel algorithm based on image similarity measures from image and video processing that maybe employed to identify change points of these networks adaptively.

Intrinsic connectivity networks (ICNs) refer to brain functional networks that are consistently found under various conditions, during tasks or at rest. Some studies demonstrated that while some stimuli do not impact intrinsic connectivity, other stimuli actually activate intrinsic connectivity through suppression, excitation, moderation or modi cation. Most analyses of functional magnetic resonance imaging (fMRI) data use ad-hoc methods to estimate the latent structure of ICNs. Modeling the effects on ICNs has also not been fully investigated. Bayesian Intrinsic Connectivity Network (BICNet) captures the ICN structure with We propose a BICNet model, an extended Bayesian dynamic sparse latent factor model, to identify the ICNs and quantify task-related effects on the ICNs. BICNet has the following advantages: (1) It simultaneously identifies the individual and group-level ICNs; (2) It robustly identifies ICNs by jointly modeling resting-state fMRI (rfMRI) and task-related fMRI (tfMRI); (3) Compared to independent component analysis (ICA)-based methods, it can quantify the difference of ICNs amplitudes across different states; (4) The sparsity of ICNs automatically performs feature selection, instead of ad-hoc thresholding. We apply BICNet to the rfMRI and language tfMRI data from the Human Connectome Project (HCP) and identify several ICNs related to distinct language processing functions.

Indiana University-Purdue University Indianapolis (IUPUI) / A popular non-invasive brain activity measurement method is based on the functional magnetic resonance imaging (fMRI). Such data are frequently used to study functional connectivity (FC) defined as statistical association among two or more anatomically distinct fMRI signals (Friston, 1994). FC has emerged in recent years as a valuable tool for providing a deeper understanding of neurodegenerative diseases and neuropsychiatric disorders, such as Alzheimer's disease and autism. Information about complex association structure in high-dimensional fMRI data is often discarded by a calculating an average across complex spatiotemporal processes without providing an uncertainty measure around it. First, we propose a non-parametric approach to estimate the uncertainty of dynamic FC (dFC) estimates. Our method is based on three components: an extension of a boot strapping method for multivariate time series, recently introduced by Jentsch and Politis (2015); sliding window correlation estimation; and kernel smoothing. Second, we propose a two-step approach to analyze and summarize dFC estimates from a task-based fMRI study of social-to-heavy alcohol drinkers during stimulation with avors. In the first step, we apply our method from the first paper to estimate dFC for each region subject combination. In the second step, we use semiparametric additive mixed models to account for complex correlation structure and model dFC on a population level following the study's experimental design. Third, we propose to utilize the estimated dFC to study the system's modularity defined as the mutually exclusive division of brain regions into blocks with intra-connectivity greater than the one obtained by chance. As a result, we obtain brain partition suggesting the existence of common functionally-based brain organization. The main contribution of our work stems from the combination of the methods from the fields of statistics, machine learning and network theory to provide statistical tools for studying brain connectivity from a holistic, multi-disciplinary perspective.

The brain connectivity of resting-state fMRI (rs-fMRI) represents an intrinsic state of brain architecture, and it has been used as a useful neural marker for detecting psychiatric conditions such as autism spectrum disorder, as well as for predicting psychosocial characteristics such as age. However, most studies using brain connectivity have focused more on the strength of functional connectivity over time (static-FC) than temporal features of connectivity changes (connectome variability). The primary goal of the current study was to investigate the effectiveness of using the connectome variability in classifying an individual’s pathological characteristics from others and predicting psychosocial characteristics. In addition, the current study aimed to prove that benefits of the connectome variability are reliable across various analysis procedures. To this end, three open public large rs-fMRI datasets including ABIDE, COBRE, and NKI were used. The static-FC and the connectome variability metrics were calculated with various brain parcellations and parameters and then utilized for subsequent machine learning (ML) classification and prediction. The results demonstrated that including the connectome variability increased the ML performances significantly in most cases of analytical variations. In addition, including the connectome variability prevented ML performance deterioration when excessive components were used. In conclusion, the current finding proved the usefulness of the connectome variability and its reliability. / M.S. / Functional magnetic resonance imaging (fMRI) with functional connectivity (FC) analysis has been widely used to understand the human brain’s system and cognitive processes. Especially, the resting-state fMRI (rs-fMRI) has been regarded as a comprehensive map of the brain’s large-scale functional architecture. Previous seminal findings demonstrated that brain regions show synchronized patterns even without any external stimulus or task (Biswal et al., 1995; Power et al., 2011), and recent studies also demonstrated that functional network architecture during tasks can be formed based on resting-state network architecture primarily suggesting that the resting-state is an intrinsic and fundamental of brain organization functionally. At the early stage of fMRI FC studies, researchers commonly adopted static measure of connectivity (static-FC) such as Pearson correlation. However, the brain has a dynamic nature, thus the static approach does not capture temporal information of the brain. In this context, time-varying or dynamic-FC has been suggested as a promising substitute. The derived dynamic-FC usually has been used to distinguish several dynamic states by identifying repeated spatial dynamic-FC profiles. Another utilization is quantifying moment-to-moment changes of dynamic-FC (connectome variability) which can represent how much dynamic-FC is stable. Interestingly, although its importance of dynamic-FC temporal features, few studies have utilized connectome variability. In addition, only a few studies compared static-FC and connectome variability (Fong et al., 2019; Wang et al., 2018). Therefore, it is necessary to demonstrate the benefits of connectome variability and its reliability across various cognitive domains and analytic procedures. To this aim, this study used three large open fMRI datasets: ABIDE comprised of autism spectrum disorder and typical development, COBRE comprised of schizophrenia and control group, and NKI which is a developmental dataset across the lifespan. In individuals’ resting-state fMRI, brain signal time series was extracted using various parcellation methods including AAL2 atlas (Rolls et al., 2015), bilateralized AAL2 atlas, and LAIRD network atlas (Laird et al., 2011). To calculate static-FC, pairwise Pearson correlation was used. For the dynamic-FC, sliding-window correlation was used with 60 second window size. Additional 90 second and 120 second sliding window sizes were also used to test the reliability of the current study. The additional sliding window sizes showed almost identical results to that of the main sliding window size (60s). The derived dynamic-FC was used to calculate ‘connectome variability’ using mean square successive difference (MSSD). The calculated static-FC and the connectome variability were inputted to support vector machine (SVM) for group classifications or support vector regression (SVR) for predicting individuals’ characteristics. Before machine learning analysis (SVM, SVR), lasso regression was adopted as a feature selection method. The SVM results showed that including connectome variability increased group classification performances in ABIDE and COBRE datasets. Interestingly, including connectome variability improved the robustness of SVM classification when the number of components was controlled. Similarly, the SVR results also demonstrated that including connectome variability increased prediction performances for autism symptom severity score (ADOS), social responsiveness score (SRS), and individuals’ age. These benefits were consistent across three parcellation schemes. In conclusion, the current study demonstrated that the connectome variability is useful to classify different groups and to predict individuals’ characteristics. Such benefits were reliable across multiple cognitive domains and robust to several analytic procedures. These results emphasized that the connectome variability which has been usually overlooked reflects some aspects of functional brain architecture, and future fMRI studies should more attend connectome variability between brain regions.

Le cerveau humain est un réseau très complexe. Le fonctionnement cérébral ne résulte donc pas de l'activation de régions cérébrales isolées mais au contraire met en jeu des réseaux distribués dans le cerveau (Bassett and Sporns, 2017; McIntosh, 2000). Par conséquent, l'analyse de la connectivité cérébrale à partir des données de neuroimagerie occupe aujourd'hui une place centrale dans la compréhension des fonctions cognitives (Sporns, 2010). Grâce à son excellente résolution spatiale, l'IRMf est devenue l'une des méthodes non invasives les plus couramment utilisées pour étudier cette connectivité. Cependant, l'IRMf a une faible résolution temporelle ce qui rend très difficile le suivi de la dynamique des réseaux cérébraux. Un défi considérable en neuroscience cognitive est donc l'identification et le suivi des réseaux cérébraux sur des durées courtes (Hutchison et al., 2013), généralement <1s pour une tâche de dénomination d'images, par exemple. Jusqu'à présent, peu d'études ont abordé cette question qui nécessite l'utilisation de techniques ayant une résolution temporelle très élevée (de l'ordre de la ms), ce qui est le cas pour la magnéto- ou l'électro-encéphalographie (MEG ou EEG). Cependant, l'interprétation des mesures de connectivité à partir d'enregistrements effectués au niveau des électrodes (scalp) n'est pas simple, car ces enregistrements ont une faible résolution spatiale et leur précision est altérée par les effets de conduction par le volume (Schoffelen and Gross, 2009). Ainsi, au cours des dernières années, l'analyse de la connectivité fonctionnelle au niveau des sources corticales reconstruites à partir des signaux du scalp a fait l'objet d'un intérêt croissant. L'avantage de cette méthode est d'améliorer la résolution spatiale, tout en conservant l'excellente résolution temporelle de l'EEG ou de la MEG (Hassan et al., 2014; Hassan and Wendling, 2018; Schoffelen and Gross, 2009). Cependant, l'aspect dynamique n'a pas été suffisamment exploité par cette méthode. Le premier objectif de cette thèse est de montrer comment l'approche « EEG connectivité source » permet de suivre la dynamique spatio-temporelle des réseaux cérébraux impliqués soit dans une tache cognitive, soit à l'état de repos. Par ailleurs, les études récentes ont montré que les désordres neurologiques sont le plus souvent associés à des anomalies dans la connectivité cérébrale qui entraînent des altérations dans des réseaux cérébraux «large-échelle» impliquant des régions distantes (Fornito and Bullmore, 2014). C'est particulièrement le cas pour l'épilepsie et les maladies neurodégénératives (Alzheimer, Parkinson) qui constituent, selon l'OMS, un enjeu majeur de santé publique. Dans ce contexte, la demande clinique est très forte pour de nouvelles méthodes capables d'identifier des réseaux pathologiques, méthodes simples à mettre en œuvre et surtout non invasives. Ceci est le deuxième objectif de cette thèse. / The human brain is a very complex network. Cerebral function therefore does not imply activation of isolated brain regions but instead involves distributed networks in the brain (Bassett and Sporns, 2017, McIntosh, 2000). Therefore, the analysis of the brain connectivity from neuroimaging data has an important role to understand cognitive functions (Sporns, 2010). Thanks to its excellent spatial resolution, fMRI has become one of the most common non-invasive methods used to study this connectivity. However, fMRI has a low temporal resolution which makes it very difficult to monitor the dynamics of brain networks. A considerable challenge in cognitive neuroscience is therefore the identification and monitoring of brain networks over short time durations(Hutchison et al., 2013), usually <1s for a picture naming task, for example. So far, few studies have addressed this issue which requires the use of techniques with a very high temporal resolution (of the order of the ms), which is the case for magneto- or electro-encephalography (MEG or EEG). However, the interpretation of connectivity measurements from recordings made at the level of the electrodes (scalp) is not simple because these recordings have low spatial resolution and their accuracy is impaired by volume conduction effects (Schoffelen and Gross, 2009). Thus, during recent years, the analysis of functional connectivity at the level of cortical sources reconstructed from scalp signals has been of increasing interest. The advantage of this method is to improve the spatial resolution, while maintaining the excellent resolution of EEG or MEG (Hassan et al., 2014; Hassan and Wendling, 2018; Schoffelen and Gross, 2009). However, the dynamic aspect has not been sufficiently exploited by this method. The first objective of this thesis is to show how the EEG connectivity approach source "makes it possible to follow the spatio-temporal dynamics of the cerebral networks involved either in a cognitive task or at rest. Moreover, recent studies have shown that neurological disorders are most often associated with abnormalities in cerebral connectivity that result in alterations in wide-scale brain networks involving remote regions (Fornito and Bullmore, 2014). This is particularly the case for epilepsy and neurodegenerative diseases (Alzheimer's, Parkinson's) which constitute, according to WHO, a major issue of public health.In this context, the need is high for new methods capable of identifying Pathological networks, from easy to use and non-invasive techniques. This is the second objective of this thesis.

L’imagerie par résonance magnétique fonctionnelle (IRMf) est un outil de choix pour cartographier d’une manière non invasive l’activité du cortex, donnant ainsi un accès à l’organisation fonctionnelle cérébrale. Cette organisation des aires cérébrales en réseaux complexes reste encore un vaste sujet d’étude, autant dans le domaine de la recherche fondamentale, pour mieux comprendre le développement et le fonctionnement du cerveau, que dans le domaine clinique, à des fins diagnostiques par exemple. Les réseaux cérébraux dits de repos, chez un sujet donné, peuvent être observés lors d’études IRMf lorsqu’aucune tâche motrice ou cognitive n’est imposée au sujet imagé. La première partie de cette thèse a permis le développement d’une méthode automatique d’identification de ces réseaux. Réalisée à l’échelle du sujet, cette méthode permet de sélectionner tous les réseaux spécifiques au sujet ce qui s’avère nécessaire dans un cadre diagnostique où l’individu prime. Au delà de la détection et de l’identification de ces réseaux, l’étude de leurs modes d’interaction dans l’espace et dans le temps et plus généralement l’analyse de la dynamique de la connectivité fonctionnelle (DCF) fait l’objet d’un intérêt grandissant. Cette analyse nécessite le développement de méthodes innovantes de traitement du signal et de l’image qui, pour l’heure, sont encore de nature exploratoire. La deuxième partie de cette thèse présente donc de nouvelles approches pour caractériser la DCF en utilisant le cadre probabiliste de modèles de Markov cachés multidimensionnels. Les mécanismes conversationnels entre réseaux cérébraux peuvent ainsi être identifiés et caractérisés à l’échelle de la seconde. Deux applications, au niveau du sujet puis du groupe, ont permis de mettre en avant les modifications des propriétés dynamiques des interactions entre réseaux sous certaines conditions ou pathologies. / The functional magnetic resonance imaging (fMRI) is a perfect tool for mapping in a non- invasive manner the activity of the cortex, giving access to the functional organization of the brain. This organization of brain areas into complex networks remains a large topic of study, both from a fundamental research perspective, to better understand the development and function of the brain, and from a clinical perspective, for diagnostic purposes for instance. The resting-state networks in a given subject can be observed in fMRI studies where no motor or cognitive tasks are imposed to the subject. The first part of this thesis focused on the development of an automatic identification method of these networks. Performed at the subject level, this method selects all the resting-state networks proper to the subject. Beyond the detection and identification of these networks, the study of interactions between these networks in space and time, and more generally the analysis of the dynamic functional connectivity (DFC), is the subject of growing interest. This analysis requires the development of innovative methods of signal or image processing that, for now, are still exploratory. The second part of this thesis thus presents new approaches to characterize the DFC using the probabilistic framework of multidimensional hidden Markov models. Conversational mechanisms between brain networks can be identified and characterized at the resolution of the second. Two applications, first on a single subject then on a group, helped to highlight the changes of dynamic properties of interaction between networks under certain conditions or diseases.

Approximately 20% of the body’s energy consumption is ongoingly consumed by the brain, where the main part is due to the neural activity, which is only increased slightly when doing a demanding task. This ongoingly neural activity are studied with the so called resting-state fMRI, which mean that the neural activity in the brain is measured for participants with no specific task. These studies have been useful to understand the neural function and how the neural networks are constructed and cooperate. This have also been helpful in several clinical research, for example have differences been identified between bipolar disorder and major depressive disorder. Recent research has focused on temporal properties of the ongoing activity and it is well known that neural activity occurs in bursts. In this study, resting-state fMRI data and temporal graph theory is used to develop a point based method (PBM) to quantify these bursts at a nodal level. By doing this, the bursty pattern can be further investigated and the nodes showing the most bursty pattern (i.e hubs) can be identified. The method developed shows a robustness regarding several different aspects. In the method is two different variance threshold algorithms suggested. One local variance threshold (LVT) based on the individual variance of the edge time-series and one global variance threshold (GVT) based on the variance of all edges time-series, where the GVT shows the highest robustness. However, the choice of threshold needs to be adapted for the aims of the current study. Finally, this method ends up in a new measure to quantify this bursty pattern named bursty centrality. The derived temporal graph theoretical measure was correlated with traditional static graph properties used in resting state and showed a low but significant correlation. By applying this method on resting-state fMRI data for 32 young adults was it possible to identify regions of the brain that showed the most dynamic properties, these regions differed between the two thresholding algorithms

It is estimated that about 35-40% of adults in the U.S. suffer from insufficient sleep. Chronic sleep deprivation has become a prevalent phenomenon because of contemporary lifestyle and work-related factors. Sleep deprivation can reduce the capabilities and efficiency of attentional performance by impairing perception, increasing effort to maintain concentration, as well as introducing vision disturbance. Thus, it is important to understand the neural mechanisms behind how chronic sleep deprivation impairs sustained attention. In recent years, more attention has been paid to the study of the integration between anatomically distributed and functionally connected brain regions. Functional connectivity has been widely used to characterize brain functional integration, which measures the statistical dependency between neurophysiological events of the human brain. Further, evidence from recent studies has shown the non-stationary nature of brain functional connectivity, which may reveal more information about the human brain. Thus, the objective of this thesis is to investigate the effects of chronic sleep deprivation on sustained attention from the perspective of dynamic functional connectivity. A modified spatial cueing paradigm was used to assess human sustained attention in rested wakefulness and chronic sleep deprivation conditions. Partial least squares approach was applied to distinguish brain functional connectivity for the experimental conditions. With the integration of a sliding-window approach, dynamic patterns of brain functional connectivity were identified in two experimental conditions. The brain was modeled as a series of dynamic functional networks in each experimental condition. Graph theoretic analysis was performed to investigate the dynamic properties of brain functional networks, using network measures of clustering coefficient and characteristics path length. In the chronic sleep deprivation condition, a compensation mechanism between highly clustered organization and ineffective adaptability of brain functional networks was observed. Specifically, a highly clustered organization of brain functional networks was illustrated with a large clustering coefficient. This organization suggested that brain utilizes more connections to maintain attention in the chronic sleep deprivation condition. A smaller impact of clustering coefficient variation on characteristics path lengths indicated an ineffective adaptability of brain functional networks in the chronic sleep deprivation condition. In the rested wakefulness condition, brain functional networks showed the small-world topology in general, with the average small-world topology index larger than one. Small-world topology was identified as an optimal network structure with the balance between local information processing and global integration. Given the fluctuating values of the index over time, small-world brain networks were observed in most cases, indicating an effective adaptability of the human brain to maintain the dominance of small-world networks in the rested wakefulness condition. On the contrary, given that the average small-world topology index was smaller than one, brain functional networks generally exhibited random network structure. From the perspective of dynamic functional networks, even though there were few cases showing small-world brain networks, brain functional networks failed to maintain the dominance of small-world topology in the chronic sleep deprivation condition. In conclusion, to the best of our knowledge this thesis was the first to investigate the effects of chronic sleep deprivation on sustained attention from the perspective of dynamic brain functional connectivity. A compensation mechanism between highly clustered organization and ineffective adaptability of brain functional networks was observed in the chronic sleep deprivation condition. Furthermore, chronic sleep deprivation impaired sustained attention by reducing the effectiveness of brain functional networks' adaptability, resulting in the disrupted dominance of small-world brain networks.

Functional brain connectivity is a marker of the brain state. Growing interest in the examination of large-scale brain network functional connectivity dynamics is accompanied by an effort to find the electrophysiological correlates. The commonly used constraints applied to spatial and spectral domains during EEG data analysis may leave part of the neural activity unrecognized. A proposed approach blindly reveals multimodal EEG spectral patterns that are related to the dynamics of the BOLD functional network connectivity. The blind decomposition of EEG spectrogram by Parallel Factor Analysis has been shown to be a useful technique for uncovering patterns of neural activity where each pattern contains three signatures (spatial, temporal, and spectral). The decomposition takes into account the trilinear structure of EEG data, as compared to the standard approaches of electrode averaging, electrode subset selection or using standard frequency bands. The simultaneously acquired BOLD fMRI data were decomposed by Independent Component Analysis. Dynamic functional connectivity was computed on the component’s time series using a sliding window correlation, and functional connectivity network states were then defined based on the values of the correlation coefficients. ANOVA tests were performed to assess the relationships between the dynamics of functional connectivity network states and the fluctuations of EEG spectral patterns. Three patterns related to the dynamics of functional connectivity network states were found. Previous findings revealed a relationship between EEG spectral pattern fluctuations and the hemodynamics of large-scale brain networks. This work suggests that the relationship also exists at the level of functional connectivity dynamics among large-scale brain networks when no standard spatial and spectral constraints are applied on the EEG data.