Radiología. 2011;53(3):236-245 ISSN: 0033-8338
RADIOLOGÍA
RADIOLOGÍA
Publicación Oficial de la Sociedad Española de Radiología Médica Incluida en Index Medicus/MEDLINE
www.elsevier.es/rx
Actividad acreditada en base a la encomienda de gestión concedida por los Ministerios de Educación, Cultura y Deporte y de Sanidad y Consumo al Consejo General de Colegios Oficiales de Médicos con 1 crédito, equivalente a 4 horas lectivas. www.seram.es
www.elsevier.es/rx
UPDATE
Magnetic resonance imaging postprocessing techniques in the study of brain connectivity M. de la Iglesia-Vayáa,h,*, J. Molina-Mateog, M.J. Escarti-Fabrab,f, L. Martí-Bonmatíc, M. Roblesa, T. Meneua, E.J. Aguilard,f, J. Sanjuáne,f a Grupo de Informática Biomédica (IBIME). Instituto de Aplicaciones de las Tecnologías de la Información y de las Comunicaciones avanzadas, Instituto ITACA, Universidad Politécnica de Valencia, Valencia, Spain b Unidad de Psiquiatría, Hospital Clínico Universitario, Valencia, Spain c Departamento de Radiología, Hospital Quirón, Valencia, Spain d Departamento de Psiquiatría, Hospital de Sagunto, Valencia, Spain e Departamento de Psiquiatría, Facultad de Medicina, Universidad de Valencia, Valencia, Spain f CIBERSAM, Centro de Investigación Biomédica en Red de Enfermedades Mentales g Centro de Biomateriales e Ingeniería Tisular, Universidad Politécnica de Valencia, Valencia, Spain h Centro de Excelencia de Imagen Biomédica (CEIB). Hospital la Fe, Conselleria de Sanitat, Valencia Spain
Received 16 June 2010; accepted 29 November 2010
KEYWORDS Brain connectivity; Functional connectivity; Effective connectivity; Connectome; Magnetic resonance imaging; Methods; fMRI; Independent component analysis; Meta-analysis
Abstract Brain connectivity is a key concept for understanding brain function. Current methods to detect and quantify different types of connectivity with neuroimaging techniques are fundamental for understanding the pathophysiology of many neurologic and psychiatric disorders. This article aims to present a critical review of the magnetic resonance imaging techniques used to measure brain connectivity within the context of the Human Connectome Project. We review techniques used to measure: a) structural connectivity b) functional connectivity (main component analysis, independent component analysis, seed voxel, meta-analysis), and c) effective connectivity (psychophysiological interactions, causal dynamic models, multivariate autoregressive models, and structural equation models). These three approaches make it possible to combine and use different statistical techniques to elaborate mathematical models in the attempt to understand the functioning of the brain. The findings obtained with these techniques must be validated by other techniques for analyzing structural and functional connectivity. This information is integrated in the Human Connectome Project where all these approaches converge to provide a representation of all the different models of connectivity. © 2010 SERAM. Published by Elsevier España, S.L. All rights reserved.
*Corresponding author. E-mail:
[email protected], delaiglesia
[email protected] (M. de la Iglesia-Vayá). 0033-8338/$ - see front matter © 2010 SERAM. Published by Elsevier España, S.L. All rights reserved.
Magnetic resonance imaging postprocessing techniques in the study of brain connectivity
PALABRAS CLAVE Conectividad cerebral; Conectividad funcional; Conectividad efectiva; Conectoma; Resonancia magnética; Métodos; RMf; ICA; Meta-analisis
237
Técnicas de análisis de posproceso en resonancia magnética para el estudio de la conectividad cerebral Resumen La noción de conectividad cerebral es un aspecto clave para entender el funcionamiento cerebral. Las metodologías para detectar y cuantificar los diferentes tipos de conectividad con técnicas de neuroimagen son en la actualidad un área de estudio fundamental en la comprensión de la fisiopatología de muchos trastornos, tanto neurológicos como psiquiátricos. Con este artículo se pretende realizar una revisión crítica de las técnicas con resonancia magnética para medir la conectividad cerebral dentro del actual contexto del proyecto Conectoma. Las técnicas revisadas se dividen en: a) conectividad estructural; b) conectividad funcional (análisis de componentes principales, análisis de componentes independientes, vóxel semilla, meta-análisis) y c) conectividad efectiva (interacciones psicofisiológicas, modelo dinámico causal, modelos autorregresivos multivariantes y modelo estructural de ecuaciones). Estos tres enfoques permiten combinar y utilizar distintas técnicas matemático-estadísticas cuyos resultados proporcionan modelos para intentar predecir la funcionalidad cerebral. Es necesario validar los hallazgos de estas técnicas con otras formas de análisis de la conectividad estructural y funcional. Esta información se integra dentro del proyecto Conectoma donde este conjunto de técnicas convergen para ofrecer una representación de todos los modelos de conectividad. © 2010 SERAM. Publicado por Elsevier España, S.L. Todos los derechos reservados.
Introduction Santiago Ramón y Cajal showed that the central nervous system (CNS) is made up of cells, like all other living tissues 1 . These cells are called neurons and their interconnections, synapses. Synapses are essential for neural functioning and communication. Much of what occurs in the CNS can be described as a network of electric currents and biochemical reactions between neurons. Since the 19 TH century, observation and study of the different syndromes and conditions caused by brain injury played an important role in the development of neurosciences. For the first time, it was possible to relate specifi c brain areas to higher mental functions, such as language and memory. However, this localized model has been superseded by current assumptions, according to which cognitive functions are not located in a specific brain area, but result from the operation of complex functional systems 2 . Thanks to functional magnetic resonance imaging (fMRI), we can relate a specific task to a certain brain activation pattern 3 , that is, a set of co-activated areas. One of the greatest challenges to present-day neuroscience is to consolidate knowledge of the patterns of brain activity. However, many neurologic and psychiatric diseases involving the CNS are not related to a focal lesion or abnormalities of a single brain area. Different disorders, such as schizophrenia and autism, are understood today as complex disorders of neural connectivity4.
In vivo neural connectivity research is one of the main objectives of neuroimaging techniques. This approach is primarily intended to improve diagnosis and the treatment of a variety of neurologic and mental disorders. This paper describes the different MR imaging techniques that have recently been developed to study brain connectivity. Advantages and limitations of these techniques, together with their possible future applications, are analyzed in this study. This information is integrated within the Connectome project, where these techniques converge to provide a representation of all the different models of connectivity. Such models are discussed below.
Brain connectivity: basic concepts There are two fundamental and complementary functional operations in the brain: segregation and integration 2. Functional segregation builds upon the assumption that certain cognitive functions are located in specifi c brain areas. Brain activity characterization in terms of functional specialization of brain areas is the main line of investigation in fMRI. However, this conception of brain activity does not reveal anything about how the different brain areas communicate with each other. In addition, the functional specialization view provides a limited account of the neural basis of much more complex processes.
238
After much research, it has been showed that these specific brain areas do not operate independently, but they interact creating dynamic neural networks that give rise to cognitive functions. As a consequence, such cognitive processes cannot be regarded as area-specific since the brain does not operate on an isolated, modular basis 5 . This notion is known as functional integration. Information transfer is the influence that the state of a system exerts on the state of an other system. It is important to have a sound knowledge of the direct connections between brain areas, including knowledge of how information is transferred between them. Each neuron is interconnected in a dynamic and complex way with a great number of other neurons through a wide variety of synaptic receptors and neurotransmitters. Neural connectivity analysis can be carried out by applying a varied set of techniques. One of them is histological and cytoarchitectonic mapping. Neurophysiologic techniques that reflect how bioelectric activity is transferred between specifi c groups of brain neurons can also be applied on animal models through microelectrode implantation. It is also feasible to analyze the differences and similarities in quantitative and temporal patterns of genic expression since connectivity is first programmed by the genes. Nevertheless, it is in vivo, non-invasive techniques, which are particularly associated with fMRI, that receive most attention in biomedicine these days. fMRI has become one of the most valuable techniques for the study of CNS connectivity thanks to the latest technological advances, particularly the generalized use of high field intensity equipments (1.5 and 4 Teslas) with a better signal and higher spatial resolution of acquired images; and the new computational models of biomedical engineering. This paper focuses on the different approaches for the study of brain connection based on MR imaging studies. These techniques are outlined in Table 1.
Structural connectivity Structural or anatomic connectivity refers to the set of physical connections between neural elements 6. The physical pattern of anatomical connections is relatively stable at short time scales (seconds to minutes), but it is likely to undergo significant morphological changes at longer time scales (days) due to plasticity of the structural connectivity patterns. At present, only invasive tracing studies provide clear evidence of direct axonal connections. These studies offer more detailed information than in vivo exploration techniques. The type of analysis for the study of structural patterns of neural connection depends on the spatial scale used. Recently developed high resolution MRI techniques with high field intensity equipments allow for in vivo examination of the human cortex at 300 mm resolution 7. This resolution is close in quality to that provided by histological analysis, so that the information obtained can be compared with cytoarchitectonic maps to achieve better MR image segmentation.
M. de la Iglesia-Vayá et al
Diffusion-weighted MR imaging. Tractography Diffusion MRI techniques are non-invasive, and allow for the study of the structural connectivity of the human brain 8. A good number of research groups showed that the analysis of the main circuits of the white matter enables 3D reconstruction of the fiber bundles, much in the same way as they are reconstructed in post-mortem dissection. However, tract tracing and diffusion tensor modeling (an algebraic model that represents the preferred diffusion direction at each voxel) are still unable to provide a fully detailed reconstruction of the main neural pathways, particularly when it comes to branched and crossed fibers 9. The methods to solve the crossing-fiber problem can only partially overcome the limitations of tensor modeling, which means that there is still much work ahead. When one single fiber bundle changes its direction within a voxel, this change opens up a variety of possible orientations of fibers that can be reconstructed through different methods. Thus, although they provide richer data than the tensor imaging model, these techniques are not the final solution. Further development of these methods seems to be necessary to clear up doubts about different interpretations of the topology of human brain white matter. Generally speaking, tractography fails to adequately measure the connection strength between neural networks. Surprisingly enough, there are few studies that compare tract reconstructions acquired through non-invasive diffusion methods with reconstructions acquired through classical neuroanatomic dissection (although there are some exceptions10,11). Direct comparisons between virtual in vivo dissections and classical neuroanatomic dissections, such as those performed by Lawes, are crucial to test the feasibility neuroimaging techniques like tractography. In short, many studies of white matter by diffusion imaging methods have been conducted ever since a n i s o t r o p i c w a t e r d i f f u s i o n w a s f i r s t o b s e r v e d 12. Nevertheless, although tractography provides us with information about structural brain connectivity, this technique fails to offer direct measurements of neural connectivity in the human brain.
Functional connectivity It should not be forgotten that when interpreting MRI data, functional connectivity is basically a statistical concept. By and large, functional connectivity links spatially remote neural networks showing some kind of interrelationship. This interrelationship is determined through statistical dependence that can be calculated by means of correlation or covariance measures. Functional connectivity is often estimated by considering all elements of a system, regardless of whether these elements are connected by direct structural links. Unlike structural connectivity, functional connectivity is highly time-dependent. It should be noted that functional connectivity does not make any explicit reference to
Magnetic resonance imaging postprocessing techniques in the study of brain connectivity
239
Table 1 Chart listing the current techniques for brain connectivity characterization. Abbreviations: DTI, Diffusion Tensor Imaging. PCA, Principal Component Analysis. ICA, Independent Component Analysis. ROI, Region of Interest. PPI, Psycho-Physiological Interactions. SCA, Seed voxel Correlation Analysis. MAR, Multivariate Autoregressive Model. GC, Granger Causality. DCM, Dynamic Causal Modeling. SEM, Structural Equation Model. Types of brain connectivity
Technique
Main feature of the technique
Limitations
Structural
Classical dissection technique Histological maps
Great anatomical accuracy Great anatomical accuracy In-vivo technique
Post mortem
DTI and Tractography
Functional
PCA ICA Group ICA, tensor ICA SCA
Effective
Patterns established according to variability Distributed and independent patterns Bivariate interaction
Post mortem It does not measure connection strength; it is not a direct method for brain connectivity estimation Under- or overestimated components Under- or overestimated components Influence on performance according to localization of the ROI Each of the studies needs to be evaluated so that they all have the same level of representativeness It only considers interactions as independent elements, which impoverishes neural system representation Exploratory method
Meta-Analysis
Accurate population inferences
PPI
Individual moderator of the two-variable connectivity
MAR – GC (Granger Causality)
Interaction between time lag and hemodynamic response function It is not a good method Powerful tool for for the design of multi-regional activity low-complexity models modeling Powerful tool for It requires knowledge multi-regional activity of neurobiologically modeling plausible models
DCM
SEM
specific directional effects (cause-effect) or to an underlying structural model13-15. Functional connectivity analysis is normally carried out through statistical methods governed by neuroimaging data (non-inferential methods), which does not reveal much information about the biological aspects underlying brain region connectivity. Functional connectivity largely relies on conventional fMRI techniques, which enables to measure variations in the signal intensity of the images. These variations are related to the hemodynamic changes involved in cellular activation, which is triggered by a specific stimulus through Blood Oxygen Level-Dependent Contrast (BOLD). Apart from this, there is an intrinsic neural activity
Brain region search
Yes Yes Yes
Yes
Yes
No
No
corresponding to spontaneous low-frequency fluctuations. This activity usually involves a group of regions engaged in the organization of the intrinsic internal activity of the brain. Most studies of these fl uctuations are conducted during the subject’s resting state, and aim to acquire fMRI dynamics in the absence of stimuli. Functional networks generated under resting-state conditions are called resting-state networks. About 60-80 % of the metabolic consumption by the brain is due to the intrinsic activity of these networks. For this reason, the study of resting-state networks is essential for accurate neural circuit description. The first studies on functional integration were based on Principal Component Analysis (PCA) to convert fMRI
240
M. de la Iglesia-Vayá et al
Figure 1 Components of interest reported in a group of schizophrenic patients in an fMRI study using an emotional auditory paradigm. These components represent the mean spatial maps representing the group. Functional connectivity is defined by a statistic. On the right is the average time series of each one of the components acquired by ICA.
data into a set of components non-correlated in time or space. Independent Component Analysis (ICA) is currently used to identify components that describe brain activity in a widely dispersed network16. PCA and ICA methods are particularly useful when the brain regions involved in a specific task or the underlying structural connectivity are unknown. The main techniques that draw on this paradigm are discussed below.
Principal component analysis The PCA method separates an image matrix into components 17 . A matrix is a two-dimensional table of numbers arranged in rows and columns that are usually used to describe linear equation systems. PCA can be performed on one subject by means of a temporal fMRI sequence or on several subjects by averaging out the components (e.g. changes of the mean value of the BOLD signal). PCA is performed in order to obtain the direction of maximum variance. This direction is regarded as the fi rst component, where spatial variance of the signal is represented by an image called eigenimage, and temporal variance across a group of subjects is represented by an associated vector called eigenvector. The subsequent variance directions are provided by the subsequent components in descending order. PCA implies converting the data into a new coordinate system defined by the maximum variance directions of the data. Eigenimages may look like standard fMRI activation maps, but the analysis is constrained by the data and they do not depend on any previous hypothesis, as is the case of the general linear model. In contrast, eigenimages capture intrinsic features of the data that are present in multiple
regions making up a wide functional network 14 . The eigenvector of each component shows how the activity of the network varies over time depending on the experimental conditions, the subject under examination or the group.
Independent component analysis ICA 18,19 is performed as a statistical model of latent variable, in other words, variables that cannot be directly observed, and thus, the idea is that n linear combinations of n independent sources are observed. Each observed variable as well as each independent component is a random variable. ICA model can be modeled as a matrix: x = As
(1)
where matrix x compiles the observed data – in this case, the fMRI images –, matrix A is the mixing matrix and matrix S represents the original sources. ICA is a generative model 20,21, which means that ICA explains how the observed data are generated by a combination of the s components. When performing ICA, only matrix x is known and matrices A and s are to be estimated based on matrix x. In the ICA model, the components are statistically independent, and have a non-Gaussian distribution. The first author who used ICA for fMRI data analysis was McKeown 22. Its applicability has opened up new avenues for the design of studies of this type of imaging. However, this applicability is not without limitations 22 . The methodology of ICA has been analyzed by different authors and there are several variations to the method 23,24.
Magnetic resonance imaging postprocessing techniques in the study of brain connectivity Peak coordinates combined across studies
Kernel convolution Density kernel 1
241
Apply significance threshold Peak density Significant or ALE map results
0 –1 –1
0
0 1 1 or ALE kernel
–1
1 0 –1 –1
0
1 1
0
–1
Figure 2 The meta-analysis includes a group analysis that combines the peaks (reported activation sites) obtained in several studies. These peaks convolve with a particular kernel (depending on the method implemented) to obtain a peak density map. This map is thresholded, resulting in a map of significant results. Modified version of Wager TD et al. Soc Cogn Affect Neurosci. 2007;2:150-8. Summary of the meta-analysis techniques (ALE and KDA).
ICA is especially well-suited to obtain intrinsic connectivity networks, such as the ones listed in Figure 1, and particularly the Default Mode Network (DMN). This network is associated with a group of regions anticorrelated with the network involved in a specific task.
Seed voxel correlation analysis Seed voxel correlation analysis (SCA) is based on the assumption that there is a coherence in spontaneous fluctuations of the BOLD signal for low frequencies25. SCA requires an a priori selection of a voxel or group of voxels (or region of interest, ROI) from which the time-series are acquired. These data are then used as a regressor in a linear correlation analysis in order to calculate, in the entire brain, the maps of functional connectivity between voxels (on a two-to-two comparison basis) that covariate with the seed region. The SCA technique has been used by many groups 26-28. The main advantage of SCA over other techniques is that SCA provides a direct answer to the following direct question: what network regions are most functionally related to the seed voxel? Thanks to this direct interpretation, SCA is more appealing to researchers than other methods.
Meta-analysis This method summarizes results across many neuroimaging studies to draw conclusions from these results in a certain knowledge field. This method can be used to establish consensus on the locations of functional regions and develop new hypotheses on structure-function correspondence. The goal of these analyses is to localize consistently activated regions (if any exist) in a body of studies dealing with the same psychological state 29. The methods essentially work by counting the number of activation peaks in each local area of brain tissue and comparing the observed number of peaks to a null-hypothesis distribution to establish a criterion for significance. The
two most representative models (Figure 2) to perform this type of analysis are the kernel density analysis (KDA) and the activation likelihood estimate (ALE). A limitation of all these functional methods is that they do not control for a large number of potentially confounding variables. Even though such variables are not directly linked to the condition under examination, they may cause changes in the activity analyzed.
Effective connectivity Effective connectivity deals with the direct influence of a brain region on the physiological activity observed in other regions 30. This paradigm is used, for instance, for the study of attention to a visual stimulus31. The analysis of effective connectivity is performed by relying on statistical models that make assumptions based on brain anatomy. This fact restricts the features of the networks to be inferred to a previously chosen subgroup of regions of interest. Therefore, these are not data-driven but hypothesis-driven methods, which are used when it is possible to determine the whole set of functional areas that were previously considered to be relevant to the specified task. Effective connectivity can be estimated by taking a time-series analysis as a reference both in stimulation paradigm studies and in resting-state networks 32 . It should be pointed out that causality extraction based on the data obtained from these time-series is sensitive to technical factors, such as the sampling velocity and the time window size 33. A list of the software applications used to perform this type of analysis is available at www. nitrc.irg 34 . The main methods that make use of this paradigm are next described.
Psychophysiological interactions Psychophysiological interactions (PPIs) are one of the simplest models for functional interaction determination based on fMRI data 35 . Given a chosen time-series (obtained from a voxel or reference region), the PPI
242
method generates whole-brain connectivity maps of such voxel, which is taken as a reference for the rest of voxels based on a regression equation. Although it is non-dynamic, the PPI model has all of the essential elements to describe a system. However, this model only considers interactions as independent elements, not as an integrated whole, which limits PPI’s capacity for neural system representation 35 . Although the PPI model is not a specific tool for system modeling, it has been shown to be particularly efficient to analyze the functional interactions of a given brain region based on the rest of brain regions and to determine causality.
Structural equation model After having been used in the social sciences for several decades, the Structural Equation Model (SEM) was introduced to neuroimaging in the early 1990s by McIntosh and Gonzalez-Lima 36 . SEM is a multivariate, hypothesis-driven technique that is based on a structural model which represents the hypothesis about the causal relations between several variables36-39. In the context of neuroimaging, these variables are the BOLD time series y 1, … y n of n brain regions and the hypothetical causal relations are based on anatomically plausible connections between the regions. The strength of each connection y i → y j is specifi ed by a so-called “path coeffi cient”, which, similarly to a partial regression coefficient, indicates how the variance of yj depends on the variance of yi if all other influences on yj are held constant. In the case of fMRI, the SEM path coeffi cients specify the system effective connectivity throughout the experimental session. In any case, it would be best to know how coupling between certain regions changes as a function of the experimentally controlled context, i.e. context-dependent changes in coupling. It must be highlighted that SEM does not take into account time sequencing because if regional time-series permuted in the same way, the estimated parameters would not change. One limitation of SEM is that researchers are restricted to use structural models of relatively low complexity since models with reciprocal connections and loops often become non-identifiable40.
Multivariate autoregressive models Autoregressive models explicitly deal with time-related aspects of causality in BOLD time-series, focusing on the causal dependence of the present on the past: each data point of a regional time series is explained as a linear combination of past data points from the same region. These MAR models represent direct influences among a set of regions whose causal interactions are inferred by their mutual predictability from past time points. Although MAR is an established statistical technique, specific implementations for fMRI were suggested only recently. A complementary MAR approach, based on the idea of ‘Granger Causality’41, was proposed by Goebel et al.42. In this framework, given two time-series y 1 and y 2 , y 1 is considered to be caused by y 2 if its dynamics can be
M. de la Iglesia-Vayá et al
predicted better using past values from y 1 and y 2 as opposed to using past values of y1 alone. This method is limited since it does not consider multiple interactions.
Dynamic causal modelling Dynamic Causal Modeling (DCM) draws on a bilinear state equation to model the neural dynamics of a cognitive system, a task that cannot be performed by fMRI. In DCM, the modeled neural activity is transformed into region-specifi c BOLD signals by a hemodynamic forward model. In the context of the standard model for fMRI analysis, this technique would be a more sophisticated procedure to specify the general linear model and convolve it with a hemodynamic response function. The main goal of DCM is to estimate parameters at the neural level so that measured and modeled BOLD signals are as similar as possible. The causal architecture of the system that we would like to identify is expressed at the level of the neuronal dynamics. Thus, what is needed to enable inferences about neural parameters are models that combine two things: (i) a simple but neurobiologically plausible model of neural dynamics; and (ii) a biophysically plausible haemodynamic forward model that describes the transformation from neural activity to BOLD. The models that meet these requirements will make it possible to fit neural and hemodynamic parameters, such that the resulting BOLD series, generated by the forward model, are optimally similar to the observed BOLD time series. DCM is the only approach to date that marries models of neural dynamics with biophysical forward models. DCM has been implemented for fMRI43.
Integrating the techniques: the Human Connectome Project The previously discussed techniques are generally used as isolated methods. Putting together the data derived from each of these methods is expected to yield integrating results offering insights into both function and behaviour. Connectome is a fundamental project to achieve this goal44,45. One of the main reasons why these connectivity techniques are not used for diagnosis in clinical settings is the fact that normal connectivity patterns are unknown. In a 2005 article45, Olaf Sporns, a neuroscience professor at Indiana University (USA), voiced concern about the lack of a standard description of the brain from an anatomical-functional viewpoint, and regarded this lack as one of the greatest problems of neuroscientific research. Brain maps today can only detect spaceoccupying lesions, cerebral infarctions or other easily identifiable gross lesions. These maps are not valid, however, when it comes to subtle and non-localized lesions. Thus, it is still necessary to produce much more detailed brain maps than those currently available. In t h e s p i r i t o f t h e H u m a n G e n o m e Pr o j e c t , S p o r n s suggested the name Connectome to refer to the project whose aim is to obtain a systematic and detailed atlas of
Magnetic resonance imaging postprocessing techniques in the study of brain connectivity
243
Figure 3 Examples of a tractographic image (left), structural connectivity (center), and functional connectivity (right). Modified version of Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey CJ, Wedeen VJ, Sporns O (2008): Mapping the structural core of human cerebral cortex. PLoS Biology 6, e159.
all brain connections (Figure 3). The National Institutes of Health (NIH) have launched this project with a view to having a healthy adult’s complete brain map by 2015. A major assumption in the Human Connectome Project is that brain regions have different functional features depending on their connections, thus establishing a close function-structure interrelationship. On this basis, structure is key to studying function because it stands to reason that neighbouring neurons process the same type of information. Connectivity between nearby brain regions will enable us to identify the patterns governing more diverse processes in different brain regions. As relatively low-cost, short-time studies of the whole human genome were inconceivable ten years ago, so it is now unthinkable that advances in bioinformatics and neuroscience will provide us with such complex patterns of human brain connectivity. If the Human Connectome Project finally succeeds, we will know which are the dysfunctional brain giving rise to a lot of neurologic and psychiatric disorders.
has a long way to go. Connectome will be particularly relevant for providing a detailed account of neural interrelations. This account will enable us to carry out a thorough and reproducible analysis of the influence that certain diseases exert on such interrelations. Finally, all of the previously described techniques have limitations. The most remarkable ones are the following: (i) in absence of clinical validation, these techniques are still at an experimental stage; (ii) the techniques analyze inter-group differences applicable to large populations of subjects, but not to isolated individuals. These techniques can only be shown to be useful for diagnosis and clinical practice if radiological prediction of anomalies is performed individually. These techniques are intended to become as reliable and specific as medical imaging in the radiological diagnosis of cerebral infarction or of multiple sclerosis. Fast advances in this field over the last years lead us to be cautiously optimistic and think that this great challenge will be overcome in the near future.
Conclusions
Authors
Over the last years, big steps have been taken towards understanding the anatomical and functional structure of the CNS. Importantly, experimental and analytic methods have been designed for characterization of neural connectivity. MR techniques developed for the structural, functional or effective characterization of brain connectivity need to be sanctioned by agreed-upon anatomical and neurophysiological procedures. The distinction between functional and effective connectivity is a necessary conceptual tool to choose the suitable methods to account for functional integration. One of the greatest challenges to present-day neuroscience is to determine the matrix of connections inside the human brain, a challenge that has been taken on by the Human Connectome Project. To date, hardly anything is known about the connection network linking up the human brain’s neural elements, so this Project still
1. 2. 3. 4. 5. 6. 7. 8. 9.
Responsible for the integrity of the study: MDIV Conception of the study: MDIV Design of the study Data collection Data analysis and interpretation Statistical analysis Bibliographic search: MDIV, JMM, MJEF Manuscript drafting: MDIV, JMM, MJEF, JS Critical review of the manuscript with intellectually relevant contributions: MDIV, LMB, MRV, TM, EJA 10. Relevants: MDIV, LMB, MRV, TM, EJA 11. Approval of the final version of the manuscript: MDIV, JMM, MJEF, LMB, MRV, TM, EJA, JS S i n c e t h i s i s a r e v i e w p a p e r, t h e n a m e s o f t h e researchers working on points 3,4,5 and 6 have not been included.
244
Funding Spanish Ministry of Health: Carlos III Health Institute, FIS02/0018, Genotypification and Genetic Psychiatry Network at the Carlos III Health Institute, G03/184, Mental Disorder Network and the CIBERSAM (Centro de Investigación Biomédica en Red de Salud Mental, Mental Health Biomedical Research Center Network) Carlos III Health Institute through the RETICS Combiomed RD07/0067/2001.
Conflict of interest The authors declare no conflict of interest.
Acknowledgments We would like to thank César Tomé and Noelia Ventura for providing bibliographic information as well as Gregorio G ó m e z a n d Ro s a Va l e n z u e l a — f r o m t h e A g e n c i a Valenciana de Salud — for supporting the functional architecture of the Human Connectome Project.
References 1. Cajal RY. Histologie du systéme nerveux de L’homme et des Vertebras. Jnl Neuropathology and experimental neurology. 1909;57:883. 2. Lee L, Harrison LM, Mechelli A. A report of the functional connectivity workshop. Dusseldorf 2002. NeuroImage. 2003; 19:457-65. 3. Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey C, Wedeen V, et al. Mapping the structural core of human cerebral cortex. PLoS Biol. 2008;6:e159. 4. Calhoun VD, Eichele T, Pearlson G. Functional brain networks in schizophrenia: a review. Front Hum Neurosci. 2009;3:1-12. 5. Mesulam M. Large-scale neurocongnitive networks and distributed processing for attention, language, and memory. Ann Neurol. 1990;8:597-613. 6. Sporns O. Graph theory methods for the analysis of neural connection patterns. In: Kötter R, editor. Neuroscience Databases. A Practical Guide. 1st ed. Düsseldorf: Springer; 2002. p. 169-83. 7. Walters NB, Eickhoff SB, Schleicher A, Zilles K, Amunts K, Egan G, et al. Observer-Independent Analysis of High-Resolution MR Images of the Human Cerebral Cortex: In Vivo Delineation of Cortical Areas. Human Brain Mapping. 2007;28:1-8. 8. Derek KJ. Studying connections in the living human brain with diffusion MRI. Cortex. 2008;52:936-52. 9. Rykhlevskaia E, Gratton G, Fabiani M. Combining structural and functional neuroimaging data for studying brain connectivity: A review. Psychophysiology. 2007;45:173-87. 10. Lawes IN, Barrica TR, Murugam V, Clark CA. Atlas based segmentation of white matter tracts of the human brain using diffusion tensor tractography and comparison with classical dissection. In International Society of Magnetic Resonance in Medicine 15th Annual Meeting, Abstract #229, Berlin, Germany, 2007.
M. de la Iglesia-Vayá et al 11. Lawes IN, Barrica TR, Murugam V, Spierings N, Evans D, Song D, et al. Atlas based segmentation of white matter traces of the human brain using diffusion tensor tractography and comparison with classical dissection. NeuroImage. 2008;39: 62-79. 12. Moseley ME, Cohen Y, Kucharczyk J. Diffusion weighted MR imaging of anisotropic water diffusion in cat central nervous system. Radiology. 1990;187:439-46. 13. Sporns O. Network Analysis. Complexity and Brain Function Complexity. 2003;8:56-60. 14. Friston KJ. Functional connectivity: the principal-component analysis of large (PET) data sets. J Cerebr Blood Flow Metab. 1993;13:5-14. 15. Friston KJ. Functional and effective connectivity in neuroimaging: A synthesis. Human Brain Mapping. 1994;2:56-78. 16. McKeown MJ, Makeig S, Brown GG, Jung TP, Kindermann SS, Bell AJ, et al. Analysis of fMRI data by blind separation into independent spatial components. Hum Brain Mapp. 1998;6: 160-88. 17. Rogers BP, Morgan VL, Newton AT, Gore JC. Assesing functional connectivity in the human brain by fMRI. Magnetic Resonance Imaging. 2007;25:1347-57. 18. Jutten C, Herault J. Blind separation of sources, part I: An adaptive algorithm based on neuromimetic architecture. Signal Processing. 1991;24:1-10. 19. Comon O. Independent component analysis. A new concept? Signal Processing. 1994;36:287-314. 20. Hyvärinen A. Independent component analysis in the presence of gaussian noise by maximizing joint likelihood. Neurocomputing. 1998;22:49-67. 21. Hyvärinen A. Fast and Robust Fixed-Point Algorithms for Independent Component Analysis. IEEE Transactions on Neural Networks. 1999;10:626-34. 22. McKeown MJ, Sejnowski TJ. Independent Component Analysis of fMRI Data: Examining the Assumptions. Human Brain Mapping. 1998;6:368-72. 23. Calhoun VD, Adali T, Pekar JJ, Pearlson GD. Latency (in) sensitive ICA. Group independent component analysis of fMRI data in the temporal frequency domain Neuroimage. 2003; 20:1661-9. 24. Beckmann CF, Smith SM. Probabilistic Independent Component Analysis for Functional Magnetic Resonance Imaging. IEEE Trans on Medical Imaging. 2004;23:137-52. 25. Biswal B, Yetkin FZ, Haughton VM, Hyde JS. Functional connectivity in the motor cortex of resting human brain using echoplanar MRI. Magn Reson Med. 1995;34:537-41. 26. Greicius MD, Krasnow B, Reiss AL, Menon V. Functional connectivity in the resting brain: a network analysis of the default mode hypothesis. Proc Natl Acad Sci U S A. 2003;100: 253-8. 27. Fox MD, Snyder AZ, Vincent JL, Corbetta M, Van Essen DC, Raichle ME. The human brain is intrinsically organized into dynamic, anticorrelated functional networks. Proc Natl Acad Sci U S A. 2005;102:9673-8. 28. Margulies DS, Kelly AM, Uddin LQ, Biswal BB, Castellanos FX, Milham MP. Mapping the functional connectivity of anterior cingulate cortex. Neuroimage. 2007;37:579-88. 29. Wager TD, Lindquist M, Kaplan L. Meta-analysis of functional neuroimaging data: current and future directions. SCAN. 2007;2:150-8. 30. Stephan KE, Friston KJ. Models of Effective Connectivity in Neural Systems. In: Jirsa Viktor K, McIntosh AR, editors.
Magnetic resonance imaging postprocessing techniques in the study of brain connectivity
31.
32.
33.
34. 35.
36.
37.
Handbook of Brain Connectivity. 1st ed. Berlin Heidelberg New Cork: Springer; 2007. p. 303-26. Friston KJ, Buechel C, Fink GR, Morris J, Rolls E, Dolan RJ. Psychophysiological and modulatory interactions in neuroimaging. Neuroimage. 1997;6:218-29. Liao W, Mantini D, Zhang Z, Pan Z, Ding J, Gong Q, Yang Y, Chen H. Evaluating the effective connectivity of resting state Networks using conditional Granger causality. Biol Cybern. 2010;102:57-69. Kaminski M, Liang H. Causal Influence: Advances in Neurosignal Analysis. Critical Reviews in biomedical Engineering. 2005;33:347-430. NITRC. The source for neuroimaging Tools and resources. http://www.nitrc.org. Published 2010 Oct 10. Friston KJ, Beuchel C, Fink GR, Morris J, Rolls E, Dolan RJ. Psychophysiological and modulatory interactions in neuroimaging. NeuroImage. 1997;6:218-29. McIntosh AR, Gonzalez-Lima F. Structural equation modeling and its application to network analysis in functional brain imaging. Hum Brain Mapping. 1994;2:2-22. Büchel C, Friston KJ. Modulation of connectivity in visual pathways by attention: cortical interactions evaluated with structural equation modelling and fMRI. Cerebral Cortex. 1997;7:768-78.
245
38. Bullmore ET, Horwitz B, Honey G, Brammer M, Williams S, Sharma T. How good is good enough in path analysis of fMRI data? NeuroImage. 2000;11:289-301. 39. Penny WD, Stephan KE, Mechelli A, Friston KJ. Modeling functional integration: a comparison of structural equation and dynamic causal models. NeuroImage. 2004;23:264-74. 40. Bollen KA. Structural equations with latent variables. 1st ed. New York: John Wiley;1989. p. 432-447. 41. Granger CWJ. Investigating causal relations by econometric models and cross-spectral methods. Econometrica. 1969;37:424-38. 42. Goebel R, Roebroeck A, Kim DS, Formisano E. Investigating directed cortical interactions in time-resolved fMRI data using vector autoregressive modeling and Granger causality mapping. Magnetic Resonance Imaging. 2003;21: 1251-61. 43. Friston KJ, Harrison L, Penny W. Dynamic causal modelling. Neuroimage. 2003;19:1273-302. 44. Tomé C. Experientia Docet. Se lo que piensas antes que tu. http://cesartomelopez.blogspot.com/2009/07/se-lo-quepiensasantes-que-tu.html. Published 2009 Jul 29. 45. Sporns O, Tononi G, Kötter R. The human connectome: A structural description of the human brain. PLoS Comput Biol. 2005;1:245-51.
