Graph Signal Processing: Overview, Challenges, and Applications

Proceedings of the IEEE, Journal Year: 2018, Volume and Issue: 106(5), P. 808 - 828

Published: April 25, 2018

Research in graph signal processing (GSP) aims to develop tools for data defined on irregular domains. In this paper, we first provide an overview of core ideas GSP and their connection conventional digital processing, along with a brief historical perspective highlight how concepts recently developed build top prior research other areas. We then summarize recent advances developing basic tools, including methods sampling, filtering, or learning. Next, review progress several application areas using GSP, analysis sensor network data, biological applications image machine

Language: Английский

---

Linking Structure and Function in Macroscale Brain Networks

Trends in Cognitive Sciences, Journal Year: 2020, Volume and Issue: 24(4), P. 302 - 315

Published: Feb. 25, 2020

The emergence of network neuroscience allows researchers to quantify the link between organizational features neuronal networks and spectrum cortical functions.Current models indicate that structure function are significantly correlated, but correspondence is not perfect because reflects complex multisynaptic interactions in structural networks.Function cannot be directly estimated from structure, must inferred by higher-order interactions. Statistical, communication, biophysical have been used translate brain function.Structure–function coupling regionally heterogeneous follows molecular, cytoarchitectonic, functional hierarchies. Structure–function relationships a fundamental principle many naturally occurring systems. However, research suggests there an imperfect connectivity brain. Here, we synthesize current state knowledge linking macroscale discuss different types assess this relationship. We argue do include requisite biological detail completely predict function. Structural reconstructions enriched with local molecular cellular metadata, concert more nuanced representations functions properties, hold great potential for truly multiscale understanding structure–function relationship central concept natural sciences engineering. Consider how conformation protein determines its chemical properties and, ultimately, folding into 3D promotes among amino acids, allowing chemically interact other molecules endowing it Conversely, disruption protein's results loss Tellingly, said denatured, highlighting idea changing has fundamentally altered nervous system analogously shaped arrangement neurons populations. synaptic projections forms hierarchy (see Glossary) nested increasingly polyfunctional neural circuits support perception, cognition, action. Modern imaging technology permits high-throughput reconstruction across spatiotemporal scales species (Box 1). Through extensive international data sharing efforts, detailed system's connection patterns available humans multiple model organisms, including invertebrate [1.Chiang A-S. et al.Three-dimensional brain-wide wiring Drosophila at single-cell resolution.Curr. Biol. 2011; 21: 1-11Abstract Full Text PDF PubMed Scopus (324) Google Scholar], avian [2.Shanahan M. al.Large-scale organization forebrain: matrix theoretical analysis.Front. Comput. Neurosci. 2013; 7: 89Crossref (111) rodent [3.Oh S.W. al.A mesoscale connectome mouse brain.Nature. 2014; 508: 207Crossref (777) Scholar,4.Bota al.Architecture cerebral association underlying cognition.Proc. Natl. Acad. Sci. U.S.A. 2015; 112: E2093-E2101Crossref (90) Scholar, primate [5.Markov N.T. weighted directed interareal macaque cortex.Cereb. Cortex. 2012; 24: 17-36Crossref (272) Scholar,6.Majka P. al.Towards comprehensive atlas connections brain: Mapping tracer injection studies common marmoset reference digital template.J. Com. Neurol. 2016; 524: 2161-2181Crossref (33) Scholar. These diagrams system, termed (SC) or connectomes, represent physical elements [7.Sporns O. al.The human connectome: description brain.PLoS 2005; 1: e42Crossref (1521) Scholar]. offers opportunity articulate functions. SC possess distinctive nonrandom attributes, high clustering short path length, characteristic small-world architecture [8.Watts D.J. Strogatz S.H. Collective dynamics networks.Nature. 1998; 393: 440Crossref (0) Populations similar tend cluster together, forming specialized modules crosslinked hub nodes diverse connectional fingerprints [9.Young M.P. systems cortex.J. Roy. Soc. Lond. B. 1993; 252: 13-18Crossref Scholar,10.Kötter R. al.Connectional characteristics areas Walker's map prefrontal cortex.Neurocomputing. 2001; 38: 741-746Crossref (20) hubs disproportionately interconnected each other, putative core [11.Hagmann al.Mapping cortex.PLoS 2008; 6: e159Crossref (2384) Scholar] 'rich club' [12.van den Heuvel al.High-cost, high-capacity backbone global communication.Proc. 109: 11372-11377Crossref (361) architectural feature potentially signals sampled integrated [13.Zamora-López G. al.Cortical form module multisensory integration on top networks.Front. Neuroinform. 2010; 4: 1PubMed Finally, spatially embedded, finite metabolic material resources [14.Bullmore E. Sporns economy organization.Nat. Rev. 13: 336Crossref (1282) resulting increased prevalence shorter, low-cost [15.Horvát S. al.Spatial embedding cost constrain layout rodents primates.PLoS 14: e1002512Crossref Scholar,16.Roberts J.A. contribution geometry connectome.NeuroImage. 124: 379-393Crossref (60) attributes replicated range tracing techniques, suggesting principles phylogeny [17.Van al.Comparative connectomics.Trends. Cogn. 20: 345-361Abstract (118) imparts distinct signature coactivation patterns. Inter-regional promote signaling synchrony distant populations, giving rise coherent dynamics, measured as regional time series electromagnetic hemodynamic activity. Systematic pairs regions can (FC) networks. Over past decade, these recorded without task instruction stimulation; 'intrinsic' ' resting-state' FC thought reflect spontaneous activity [18.Biswal al.Functional motor cortex resting using echo-planar MRI.Magn. Reson. Med. 1995; 34: 537-541Crossref (5503) Intrinsic highly organized [19.Damoiseaux J. al.Consistent resting-state healthy subjects.Proc. 2006; 103: 13848-13853Crossref (2665) 20.Bellec al.Multi-level bootstrap analysis stable clusters fMRI.NeuroImage. 51: 1126-1139Crossref (170) 21.Thomas Yeo intrinsic connectivity.J. Neurophysiol. 106: 1125-1165Crossref (2040) reproducible [22.Gordon E.M. al.Precision mapping individual brains.Neuron. 2017; 95: 791-807Abstract (140) Scholar,23.Noble decade test-retest reliability connectivity: systematic review meta-analysis.NeuroImage. 2019; : 116157Crossref (5) comparable task-driven [24.Smith S.M. al.Correspondence brain's during activation rest.Proc. 2009; 13040-13045Crossref (2684) Scholar,25.Cole M.W. al.Intrinsic task-evoked architectures brain.Neuron. 83: 238-251Abstract (516) persistent nature rest makes ideal starting point study [26.Honey C.J. al.Can brain?.NeuroImage. 52: 766-776Crossref (291) Scholar,27.Damoiseaux J.S. Greicius M.D. Greater than sum parts: combining connectivity.Brain Struct. Funct. 213: 525-533Crossref Here first show direct one-to-one links limited inherently obscured networked survey modern quantitative methods move away correlations conceptualizing emerging focus strengths, limitations, commonalities. posit next steps network-level take account heterogeneity enriching microscale transcriptomic, neuromodulatory information. close theories uniform brain, vary parallel cytoarchitectonic representational Early emphasized weights. weights correlated [28.Honey C. al.Predicting connectivity.Proc. 2035-2040Crossref (1543) also Furthermore, structurally connected display greater unconnected Scholar,29.Shen K. al.Information processing functionally defined 32: 17465-17476Crossref (63) Scholar (Figure 1A). More globally, networks, particularly visual somatomotor circumscribed dense anatomical [29.Shen 30.Van Den al.Functionally linked brain.Hum. Brain Mapp. 30: 3127-3141Crossref (625) 31.Alves P.N. al.An improved neuroanatomical default-mode reconciles previous neuroimaging neuropathological findings.Commun. 2: 1-14Crossref While perfect. Even best-case estimates place correlation R2 ≈ 0.5 which means considerable variance (at least half) unexplained simple 1:1 structure. discrepancy widens case 1B ). A salient example homotopic corresponding structures two hemispheres. typically strongest subset [32.Mišić landscape One. 9: e111007Crossref (14) all supported callosal projection [33.Shen al.Stable long-range interhemispheric coordination projections.Proc. 6473-6478Crossref (52) strong may observed even individuals no [34.Uddin L.Q. al.Residual split-brain revealed fMRI.Neuroreport. 19: 703Crossref (96) 35.O'Reilly J.X. al.Causal effect disconnection lesions rhesus monkeys.Proc. 110: 13982-13987Crossref (106) 36.Layden E.A. al.Interhemispheric zebra finch absent corpus callosum normal ontogeny.NeuroImage. Crossref (1) examples illustrate sustained communication via indirect manifest FC. discordance pronounced mesoscopic scale. commonly meta-analytic recovered [37.Mišić al.Cooperative competitive spreading connectome.Neuron. 86: 1518-1529Abstract Scholar,38.Betzel R.F. al.Diversity meso-scale non-human connectomes.Nat. Commun. 2018; 346Crossref (21) 1C). reproducibly independent component community detection [39.Power J.D. 72: 665-678Abstract (1499) data-driven [20.Bellec Scholar,21.Thomas both recordings application diffusion-weighted covariance yields contiguous Scholar,40.Betzel modular networks: accounting wiring.Net. 42-68Crossref For example, fail identify default mode-like network, perhaps parts anatomically inter-connected differences. evidence assortative mixing, whereby (e.g., degrees) likely connected, whereas same true [50.Lim al.Discordant two-layer multiplex network.Sci. Rep. 2885Crossref (2) At scale, communities assortative, while disassortative [38.Betzel In words, affinity dissimilar attributes. As result, tuning algorithms sensitive improves match Altogether, rich body work demonstrates spans scales, edges their arrangement. Why FC? Functional arise connections, courses synapses removed other. propensity correlate driven only them, inputs they receive sensory organs entire [27.Damoiseaux Scholar,51.Bettinardi R.G. al.How sculpts function: unveiling structure.Chaos. 27: 047409Crossref (12) corollary much less distance-dependent connections. Anatomical subject material, spatial, constraints Scholar]; pressures reduced probability weight increasing spatial separation Although distance-dependence FC, weaker, ensuring differences configurations. section consider emergent property links. seen so far, exists nontrivial perfectly aligned. number emerged embody link, statistical [41.Mišić al.Network-level structure-function neocortex.Cereb. 26: 3285-3296Crossref (153) Scholar,42.Messé A. al.Relating relative contributions anatomy, stationary non-stationarities.PLoS. 10: e1003530Crossref Scholar,43.Graham D. Rockmore packet switching brain.J. 23: 267-276Crossref (18) 44.Goñi al.Resting-brain predicted analytic measures 111: 833-838Crossref (208) 45.Crofts J.J. Higham communicability measure applied networks.J. Interf. 411-414Crossref (61) [46.Honey al.Network shapes scales.Proc. 2007; 104: 10240-10245Crossref (941) 47.Breakspear Dynamic large-scale activity.Nat. 340Crossref (147) 48.Sanz-Leon al.Mathematical framework modeling Virtual Brain.NeuroImage. 385-430Crossref 49.Deco al.Key role coupling, delay, noise fluctuations.Proc. 10302-10307Crossref (372) Though implementation assumptions, emphasize collective, transcends geometric dependence dyadic relationships. briefly strategies, interpretation predictive utility, most importantly, what teach us about Perhaps simplest way statistically. Varying rank regression useful, canonical [52.Deligianni F. al.NODDI tensor-based microstructural indices predictors connectivity.PLoS 11: e0153404Crossref (13) partial squares objective simultaneously combinations maximally [53.McIntosh A.R. Mišić Multivariate analyses data.Annu. Psychol. 64: 499-525Crossref (73) 2). An appealing such modes. particular configuration subnetwork give Taking further, artificial learn recent variant word2vec algorithm build low-dimensional representation train deep edge-wise [54.Rosenthal relations embedded vector 2178Crossref (3) offer associate assuming specific mode interaction Communication science telecommunication engineering conceptualize superposition elementary events [43.Graham Scholar,55.Avena-Koenigsberger al.Communication networks.Nat. 17Crossref (92) By explicitly formulating inter-regional signaling, open important questions, namely: biologically realistic model, well does fit network? focused centralized shortest routing, discrete travel set source node prespecified target node. recently, attention shifted decentralized mechanisms where diffuse through [56.Mišić convergence zone hippocampus.PLoS. e1003982Crossref Scholar,57.Atasoy al.Human connectome-specific harmonic waves.Nat. 10340Crossref often broadcast fronts Scholar,58.Abdelnour diffusion accurately networks.NeuroImage. 90: 335-347Crossref (71) Scholar,59.Worrell J.C. al.Optimized sensory-motor integration.Net. 415-430Crossref Others considered neither fully nor decentralized, ensembles [45.Crofts Scholar,60.Avena-Koenigsberger al.Path tradeoff efficiency resilience connectome.Brain 222: 603-618Crossref (17) multiplexed strategies involving [44.Goñi Scholar,61.Avena-Koenigsberger routing networks.PLoS 15: e1006833Crossref 62.Betzel al.Structural, genetic factors interregional probed electrocorticography.Nat. Biomed. Eng. 63.Vazquez-Rodriguez al.Gradients tethering neocortex.Proc. 116: 21219-21227Crossref (6) consensus that, given topological proximity possible utilize either al.Re

Language: Английский

---

Gradients of structure–function tethering across neocortex

Proceedings of the National Academy of Sciences, Journal Year: 2019, Volume and Issue: 116(42), P. 21219 - 21227

Published: Sept. 30, 2019

The white matter architecture of the brain imparts a distinct signature on neuronal coactivation patterns. Interregional projections promote synchrony among distant populations, giving rise to richly patterned functional networks. A variety statistical, communication, and biophysical models have been proposed study relationship between structure function, but link is not yet known. In present report we seek relate structural connection profiles individual areas. We apply simple multilinear model that incorporates information about spatial proximity, routing, diffusion regions predict their connectivity. find structure–function relationships vary markedly across neocortex. Structure function correspond closely in unimodal, primary sensory, motor regions, diverge transmodal cortex, particularly default mode salience divergence systematically follows cytoarchitectonic hierarchies. Altogether, results demonstrate networks do align uniformly brain, gradually uncouple higher-order polysensory

Language: Английский

---

Cognitive and behavioural flexibility: neural mechanisms and clinical considerations

Nature reviews. Neuroscience, Journal Year: 2021, Volume and Issue: 22(3), P. 167 - 179

Published: Feb. 3, 2021

Cognitive and behavioural flexibility permit the appropriate adjustment of thoughts behaviours in response to changing environmental demands. Brain mechanisms enabling have been examined using non-invasive neuroimaging approaches humans alongside pharmacological lesion studies animals. This work has identified large-scale functional brain networks encompassing lateral orbital frontoparietal, midcingulo-insular frontostriatal regions that support across lifespan. Flexibility can be compromised early-life neurodevelopmental disorders, clinical conditions emerge during adolescence late-life dementias. We critically evaluate evidence for enhancement through cognitive training, physical activity bilingual experience. is critical optimal adaptation actions under circumstances. In this Review, Uddin summarizes research processes neural systems supporting discusses ways improve

Language: Английский

---

Development of structure–function coupling in human brain networks during youth

Proceedings of the National Academy of Sciences, Journal Year: 2019, Volume and Issue: 117(1), P. 771 - 778

Published: Dec. 24, 2019

The protracted development of structural and functional brain connectivity within distributed association networks coincides with improvements in higher-order cognitive processes such as executive function. However, it remains unclear how white-matter architecture develops during youth to directly support coordinated neural activity. Here, we characterize the structure-function coupling using diffusion-weighted imaging n-back MRI data a sample 727 individuals (ages 8 23 y). We found that spatial variability aligned cortical hierarchies specialization evolutionary expansion. Furthermore, hierarchy-dependent age effects on localized transmodal cortex both cross-sectional subset participants longitudinal (n = 294). Moreover, rostrolateral prefrontal was associated performance partially mediated age-related Together, these findings delineate critical dimension adolescent development, whereby between remodels cognition.

Language: Английский

---

The physics of brain network structure, function and control

Nature Reviews Physics, Journal Year: 2019, Volume and Issue: 1(5), P. 318 - 332

Published: March 27, 2019

Language: Английский

---

Decoupling of brain function from structure reveals regional behavioral specialization in humans

Nature Communications, Journal Year: 2019, Volume and Issue: 10(1)

Published: Oct. 18, 2019

The brain is an assembly of neuronal populations interconnected by structural pathways. Brain activity expressed on and constrained this substrate. Therefore, statistical dependencies between functional signals in directly connected areas can be expected higher. However, the degree to which function bound underlying wiring diagram remains a complex question that has been only partially answered. Here, we introduce structural-decoupling index quantify coupling strength structure function, reveal macroscale gradient from regions more strongly coupled, decoupled, than realistic surrogate data. This spans behavioral domains lower-level sensory high-level cognitive ones shows for first time structure-function spatially varying line with evidence derived other modalities, such as connectivity, gene expression, microstructural properties temporal hierarchy.

Language: Английский

---

Stationary Graph Processes and Spectral Estimation

IEEE Transactions on Signal Processing, Journal Year: 2017, Volume and Issue: 65(22), P. 5911 - 5926

Published: Aug. 11, 2017

Stationarity is a cornerstone property that facilitates the analysis and processing of random signals in time domain. Although time-varying are abundant nature, many practical scenarios information interest resides more irregular graph domains. This lack regularity hampers generalization classical notion stationarity to signals. The contribution this paper twofold. Firstly, we propose definition weak for takes into account structure where process place, while inheriting meaningful properties Our requires stationary processes can be modeled as output linear filter applied white input. We will show equivalent requiring correlation matrix diagonalized by Fourier transform. Secondly, analyze power spectral density number methods estimate it. start with nonparametric approaches, including periodograms, window-based average banks. then shift focus parametric discussing estimation moving-average (MA), autoregressive (AR) ARMA processes. Finally, illustrate synthetic real-world graphs.

Language: Английский

---

A Graph Signal Processing Perspective on Functional Brain Imaging

Proceedings of the IEEE, Journal Year: 2018, Volume and Issue: 106(5), P. 868 - 885

Published: March 7, 2018

Modern neuroimaging techniques provide us with unique views on brain structure and function; i.e., how the is wired, where when activity takes place. Data acquired using these can be analyzed in terms of its network to reveal organizing principles at systems level. Graph representations are versatile models nodes associated regions edges structural or functional connections. Structural graphs model neural pathways white matter, which anatomical backbone between regions. Functional built based connectivity, a pairwise measure statistical interdependency pairs regional traces. Therefore, most research date has focused analyzing reflecting function. signal processing (GSP) an emerging area signals recorded graph studied atop underlying structure. An increasing number fundamental operations have been generalized setting, allowing analyze from new viewpoint. Here, we review GSP for imaging data discuss their potential integrate structure, contained itself, function, residing signals. We meaningfully filtered concepts spectral modes derived also derive other such as surrogate generation decompositions informed by cognitive systems. In sum, offers novel framework analysis data.

Language: Английский

---

Heritability and interindividual variability of regional structure-function coupling

Nature Communications, Journal Year: 2021, Volume and Issue: 12(1)

Published: Aug. 12, 2021

Abstract White matter structural connections are likely to support flow of functional activation or connectivity. While the relationship between and connectivity profiles, here called SC-FC coupling, has been studied on a whole-brain, global level, few studies have investigated this at regional scale. Here we quantify coupling in healthy young adults using diffusion-weighted MRI resting-state data from Human Connectome Project study how may be heritable varies individuals. We show that strength widely across brain regions, but was strongest highly structurally connected visual subcortical areas. also interindividual differences based age, sex composite cognitive scores, within certain networks. These results suggest structure-function is an idiosyncratic feature organisation influenced by genetic factors.

Language: Английский

---