Functional Imaging

From Scholarpedia
William D. Penny and Karl J. Friston (2007), Scholarpedia, 2(5):1478. doi:10.4249/scholarpedia.1478 revision #91282 [link to/cite this article]
(Redirected from Brain imaging)
Jump to: navigation, search

Functional imaging is the study of human brain function based on analysis of data acquired using brain imaging modalities such as Electroencephalography (EEG), Magnetoencephalography (MEG), functional Magnetic Resonance Imaging (fMRI), Positron Emission Tomography (PET) or Optical Imaging. The aim is to understand how the brain works, in terms of its physiology, functional architecture and dynamics. The framework for the conduct of these studies includes classical techniques of neuroanatomy, neurophysiology, and experimental psychology and the cognitive neurosciences, as well as more theoretical approaches, based on perspectives from computational neuroscience and statistics.

Modern functional imaging has two main advantages over the multi/single-unit recordings used to study the electrophysiology of neurons. The first is that it is generally non-invasive, and is therefore applicable routinely in humans. This allows for the study of unique human attributes such as language. The second is that it can provide a wide field of view. Rather than recording information about a single or small number of neuronal cells, an image may be gathered summarizing simultaneous activity across the whole brain. This provides a different yet complementary perspective on neural coding (see e.g., functional integration, below). A disadvantage, however, is that functional imaging provides only an indirect measure of the quantities of primary interest to neuroscientists e.g., firing rates and membrane potentials. Current research is aimed at bridging this gap using a combination of experimental and mathematical modelling approaches.

Contents

Modalities

Current imaging modalities include the Electroencephalogram (EEG) which records electrical voltages from electrodes placed on the scalp and the Magnetoencephalogram (MEG) which records the magnetic field from SQUID sensors placed above the head. Both MEG and EEG have a high temporal resolution (milliseconds), capable of detecting e.g., the 40Hz Gamma response implicated in object representation (Tallon-Baudry and Bertrand, 1999). Their spatial resolution is, however, usually of the order of centimeters rather than millimeters. This varies a great deal, depending on the nature of the neuronal activity one is trying to localize. It depends in particular on the number of sources that is activated at the time data is recorded. In practice this implies that e.g. for isolating subtle cognitive components, a lower resolution is to be expected, whilst the stronger early components of an auditory response can be localized to within millimeters in the brainstem.

In contrast, functional Magnetic Resonance Imaging (fMRI) has low temporal (hundreds of milliseconds or seconds) but relatively high spatial (millimeters) resolution. Increases in neural activity cause variations in blood oxygenation, which in turn cause magnetization changes that can be detected in an MRI scanner. This Blood Oxygenation Level Dependent (BOLD) signal peaks up to 6s after neuronal activity. Moreover, the hemodynamics act like a low pass filter (Logothetis, 2001), smearing out changes in local electrical activity.

Simultaneous recordings of EEG and fMRI (Ritter and Villringer, 2006) have the potential to localise neuronal activity with both high temporal and spatial resolution. Other important imaging modalities are PET and Optical Imaging. PET's spatial resolution typically falls somewhere between that of fMRI and MEG/EEG. In addition, PET has very low temporal resolution (tens of seconds to minutes) and requires injection of a trace amount of radioactivity. This limits the number of measurements that can be made on any one individual. But a great advantage of PET is that it is particularly useful in the study of brain neurophysiology and neurochemistry e.g., one can image glucose uptake and the activity at serotonin and dopamine receptors, in systems of importance to those studying anxiety, depression and addiction. Optical Imaging [1] or Near Infrared Spectroscopy (NIRs) can also detect BOLD signals from changes in the amount of reflected light. This is an economical alternative to fMRI but is limited to imaging the cortex.

Functional imaging is also closely related to structural imaging, in which MRI is used to provide high resolution images with high contrast between e.g., white matter and gray matter. These detailed anatomical images have recently been complemented with data from Diffusion Tensor Imaging (DTI) which can show the direction of white matter fibres.

Functional Specialization

There are two key themes in the analysis of functional imaging data. They reflect the long-standing debate in neuroscience about functional specialization versus functional integration in the brain (Cohen and Tong, 2001). The first is brain `mapping’ where three-dimensional images of neuronal activation are produced showing which parts of the brain respond to a given cognitive or sensory challenge. This is also known as the study of functional specialization and generally proceeds using some form of Statistical Parametric Mapping (SPM). A classic example here is the identification of human V4 and V5, the areas specialized for the processing of color and motion.

SPM is a voxel-based approach, employing classical statistics and topological inference, to make comments about regionally specific responses to experimental factors. PET or fMRI data are first spatially processed so that they conform to a known anatomical space, in which responses are characterized statistically typically using the General Linear Model (GLM). For fMRI data the GLM embodies a convolution model of the hemodynamic response. This accounts for the fact that BOLD signals are a delayed and dispersed version of the neuronal response. GLMs are fitted at each voxel and inferences are made about which parts of the brain are active, in a statistical sense. To accommodate the spatial nature of the imaging data (and account for the multiple statistical comparisons made) SPM techniques make use of Random Field Theory (RFT) (see Fig 1) and/or other statistical procedures, e.g., False Discovery Rate.

Figure 1: Processing stream for brain mapping using PET or fMRI data

The SPM approach can also be used with structural data to find brain regions containing a higher gray matter density. This is known as Voxel-Based Morphometry (VBM) (Ashburner and Friston, 2000) and has been used, for example, to show that the posterior hippocampus, useful for spatial navigation, is enlarged in taxi drivers.

For MEG or EEG, data can be analyzed in sensor space, furnishing a crude spatial mapping of brain function. Functions can, however, be more accurately localized using source reconstruction methods (Baillet et al. 2001). These work by specifying a forward model describing how a current source in the brain propagates to become an MEG or EEG measurement, using Maxwell's equations (http://www.scholarpedia.org/article/Volume_Conduction). These models are then inverted using statistical inference. Data from sensory systems is often analyzed using an averaging procedure. The data immediately following a sensory event, e.g., hearing an auditory tone, is averaged over multiple events to produce an Event Related Potential (ERP). Components of the ERP can then be localized to different parts of the brain. Other cognitive components, however, are not easily isolated using this ERP approach. For these, a time-frequency characterization may be more appropriate (Tallon-Baudry and Bertrand, 1999). See also Makeig et al. 2002 for a recent critique of the averaging procedure.

Figure 2: Sensor distribution of EEG at a single time point and ERPs for two different experimental conditions

Functional Integration

The second theme is ‘functional integration’, where models are used to describe how different brain areas interact. A classic example is the use of models to find increased connectivity between dorsal and ventral visual streams after subjects learn object-place associations. A wide range of statistical techniques are being used to measure inter-regional connectivity. Both unsupervised (e.g., Independent Component Analysis , ICA) and supervised techniques (e.g., support vector machine, SVM) are used. Other models seek to directly measure "causal" connectivity based on static, statistical constraints (e.g., Structural Equation Modelling, SEM) or dynamic, more bio-physically motivated assumptions (e.g., Dynamic Causal Modelling, DCM). A challenge for functional integration models is to bridge the gap between the large-scale, statistical models of the whole brain, and the small number of highly constrained spatial regions needed to be able to apply SEM and/or DCM.

DCM for fMRI uses a forward model in which neural activity generates BOLD signal changes via a `Balloon' model of vascular dynamics. The model is then inverted to provide estimates of changes in connectivity between brain regions. In DCM for ERPs, neural activity is described using neural-mass models, which then give rise to observed EEG or MEG data using Maxwell's equations (see above). Inversion of the model then allows one to make inferences about changes in long-range excitatory connections among different brain areas.

The above analysis approaches are implemented in various software packages such as SPM [2] (SPM is the name of a software package as well as a methodology), FSL [3], EEGLAB [4], BrainVoyager [5], or AFNI [6]. They are also described in a recent textbook (Friston et al. 2006).

Applications

The applications of functional imaging are diverse and multitudinous. PubMed, for example, returns over 32,000 articles. The functional imaging journal `NeuroImage' classifies research articles under 'Anatomy and Physiology', 'Methods and Modelling', 'Systems Neuroscience' or 'Cognitive Neuroscience'. Additionally a number of applications in clinical and experimental medicine are emerging.

Functional imaging has been applied to all systems of the brain; whether visual, auditory, sensorimotor, emotional, memory, language, attention or control. Overviews of research findings are available in recent textbooks (Frackowiak et al. 2003, Gazzaniga 2004). Recent high-profile (and arbitrarily selected) applications of fMRI in these areas include a study of the effect of sleep on human memory performance (Yoo et al. 2007) and a study of the neuronal and cognitive components of altruism (Tankersley et al. 2007).

Imaging is also used for the study of basic brain anatomy and physiology. For example, DTI has recently been used to identify three regions of human parietal cortex based on their connectivity patterns with other brain areas (Rushworth et al. 2006). Imaging is also used clinically: The best established application is the use of fMRI for pre-surgical mapping to localize cerebral functions in tissue within or near regions intended for neurosurgical resection (Matthews et al. 2006).

Authors web page: http://www.fil.ion.ucl.ac.uk/~wpenny/

Recommended Reading

R.S.J. Frackowiak, K.J. Friston, C. Frith, R. Dolan, C.J. Price, S. Zeki, J. Ashburner, and W.D. Penny (2003) Human Brain Function. Academic Press, UK, 2nd edition.

C. Frith (2007) Making up the mind: How the brain creates our mental world. Blackwell Publishing.

K. Friston, J. Ashburner, S. Kiebel, T. Nichols and W. Penny (2006) Statistical Parametric Mapping: The Analysis of Functional Brain Images. Elsevier, London.

Gazzaniga, M.S., Ivry, R., & Mangun, G.R. Cognitive Neuroscience: The Biology of the Mind. W.W. Norton, 2002. 2nd Edition

M.S. Gazzaniga (2004). The Cognitive Neurosciences III. MIT Press, New York.

Internal references

References

J. Ashburner and K.J. Friston. Voxel-Based Morphometry - The Methods. NeuroImage, 11:805-821, 2000

S. Baillet, J.C. Mosher and R.M. Leahy (2001) Electromagnetic brain mapping. IEEE Signal Processing Magazine, pages 14-30.

J.D. Cohen and F. Tong (2001) The face of controversy. Science, 293, 2405-2407.

N. Logothetis, J. Pauls, M. Augath, T. Trinath and A. Oeltermann (2001) Neurophysiological investigation of the basis of the fMRI signal. Nature 412, 150-157.

S. Makeig, M. Westerfield, T Jung, S. Enghoff, J. Townsend, E Courchesne and T. Sejnowski (2002) Dynamic brain sources of visual evoked responses Science, 295, 690-694.

P.M. Matthews, G.D. Honey and E.T. Bullmore (2006) Applications of fMRI in translational medicine and clinical practice. Nature Reviews Neuroscience, 7, 732-744.

M.F. Rushworth, T.E. Behrens and H. Johansen-Berg (2006) Connection patterns distinguish 3 regions of human parietal cortex. Cerebral Cortex, 16(10):1418-30.

C. Tallon-Baudry and O. Bertrand (1999) Oscillatory gamma activity and its role in object representation. Trends in Cognitive Sciences, 3(4), 151-162.

P. Ritter and A. Villringer (2006) Simultaneous EEG-fMRI. Neuroscience and Biobehavioural Reviews. 30(6), 823-838.

D. Tankersley, C J Stowe and S A Huettel (2007) Altruism is associated with an increased neural response to agency. Nature Neuroscience 10, 150 - 151.

S.S. Yoo, P T Hu, N. Gujar, F A Jolesz and M P Walker (2007). A deficit in the ability to form new human memories without sleep. Nature Neuroscience 10, 385 - 392.

External Links

http://www.fil.ion.ucl.ac.uk/spm/

http://www.fmrib.ox.ac.uk/fsl/

http://www.sccn.ucsd.edu/eeglab/

http://afni.nimh.nih.gov/afni/

http://www.humanbrainmapping.org/

http://www.elsevier.com/wps/find/journaldescription.cws_home/622925/description#description

http://www3.interscience.wiley.com/cgi-bin/jhome/38751

See also

Computational Neuroanatomy, Current Source Density, Diffusion Tensor Imaging, Event-Related Brain Dynamics, fMRI, MEG, MRI, Neural Networks, Transcranial Magnetic Stimulation, Neurovascular Coupling

Personal tools
Namespaces

Variants
Actions
Navigation
Focal areas
Activity
Tools