What does fMRI tell us about neuronal activity?

Key Points

  • Functional magnetic resonance imaging (fMRI) is now widely used in cognitive neuroscience to look for changes in neural activity that correlate with particular cognitive processes. But what does fMRI tell us? Until we know exactly what the fMRI signal represents, we will not know how to interpret the results of these studies.

  • It is generally assumed that the fMRI signal is roughly proportional to a measure of local neural activity, averaged over several millimetres and several seconds — the linear transform model of the fMRI signal. The relationship between the signal and the underlying neural activity depends on the fMRI acquisition technique, the behavioural and stimulation protocol, and how neural activity is measured and quantified (for example, by firing rate, local field potential or synchronous firing).

  • The linear transform model predicts that the fMRI response should sum over time. For example, the fMRI response to a 12-s stimulus should be the same as that to two 6-s stimuli. The frequently observed failure of this prediction could be due to a ceiling effect, or to the disproportionately large responses to very short stimuli that are seen in primary sensory cortex.

  • It should be possible to predict the fMRI response from the underlying neural activity. Studies that have compared neural activity and fMRI responses indirectly have supported this relationship. A direct comparison of simultaneously recorded fMRI responses and neuronal signals in monkey visual cortex found that local field potentials predicted fMRI responses better than multi-unit activity, but that both were variable in accuracy.

  • The linear transform model also predicts that fMRI responses and neural activity should be colocalized. There is increasing evidence that this is the case, although the analysis methods used influence the localization of the fMRI signal.

  • It will be important to improve our understanding of the relationships between neural activity, blood flow and metabolism, and fMRI signals, if we are correctly to interpret fMRI studies. It will also be necessary to optimize protocols and analysis techniques if this powerful tool is to live up to its potential.

Abstract

In recent years, cognitive neuroscientists have taken great advantage of functional magnetic resonance imaging (fMRI) as a non-invasive method of measuring neuronal activity in the human brain. But what exactly does fMRI tell us? We know that its signals arise from changes in local haemodynamics that, in turn, result from alterations in neuronal activity, but exactly how neuronal activity, haemodynamics and fMRI signals are related is unclear. It has been assumed that the fMRI signal is proportional to the local average neuronal activity, but many factors can influence the relationship between the two. A clearer understanding of how neuronal activity influences the fMRI signal is needed if we are correctly to interpret functional imaging data.

Main

Functional magnetic resonance imaging (fMRI)1,2,3,4,5 has revolutionized cognitive neuroscience over the past decade. fMRI takes advantage of the coupling between neuronal activity and haemodynamics (the local control of blood flow and oxygenation) in the brain to allow the non-invasive localization and measurement of brain activity (Box 1). This technique is allowing a new era of research, complementary to more invasive techniques for measuring neuronal activity in animal models, to explore the function and dysfunction of the human brain. However, the ultimate success of this research depends on the relationship between the fMRI signal and the underlying neuronal activity. The vascular source of the fMRI signal places important limits on the usefulness of the technique. Although we know that the fMRI signal is triggered by the metabolic demands of increased neuronal activity, the details of this process are only partially understood (Box 2). Consequently, this issue has emerged as one of the most important areas in neuroscience.

Understanding the relationship between fMRI and neuronal responses is particularly important because some fMRI and neurophysiology experiments have yielded apparently conflicting results. One example concerns the role of primary visual cortex (V1) in spatial attention6. Our ability to perform a visual discrimination task improves when we are cued to attend, without moving our eyes, to the spatial location of the relevant stimulus. Shifts in attention are correlated with systematic changes in activity in a number of brain areas. Although some theories posit that attention is mediated by selection very early in the visual pathways, attentional effects have been notoriously difficult to measure using single-unit electrophysiology in area V1 of the monkey brain7,8. By contrast, robust attentional effects have been readily measurable with fMRI in human V1, and attentional effects in EXTRASTRIATE CORTICAL AREAS (such as V4) seem to be considerably larger in human fMRI experiments than in monkey electrophysiology experiments9,10. Is this due to a species difference, or does it indicate a difference in what is being measured? For example, the fMRI signal might reflect not only the firing rates of the local neuronal population, but also subthreshold activity, simultaneous excitation and inhibition, or modulatory inputs (such as feedback from distant, higher-level areas of visual cortex) that might not evoke spikes. In addition, the fMRI signal might reflect changes in neuronal synchrony without a concomitant increase in mean firing rate11. A further complication is that fMRI signals reflect the pooled activity of a very large number of neurons; modulations in the fMRI responses could be caused by either large changes in the firing rates in a small subpopulation of neurons, or small changes in the firing rates in a much larger subpopulation of neurons12.

In this review, we examine our current understanding of how the fMRI signal relates to neuronal responses, in the context of a simple linear transform model. We begin by defining the model, and then discuss experiments that have evaluated three crucial implications of the model: that fMRI responses should show temporal summation, that the time course and amplitude of the fMRI responses should be predictable from the underlying neuronal responses, and that the fMRI responses should be colocalized with the underlying neuronal activity.

The linear transform model

The central assumption guiding inferences that are made from fMRI data about neuronal activity has been that the fMRI signal is approximately proportional to a measure of local neural activity, averaged over a spatial extent of several millimetres and over a time period of several seconds13,14 (Fig. 1). We will refer to this as the linear transform model of the fMRI signal. Note that the linear transform model allows the possibility of an arbitrarily complex, nonlinear relationship between the stimulus and the neuronal activity. To evaluate the model, one must either measure the underlying neuronal activity (as in the direct comparisons of fMRI and electrophysiological measurements discussed below), or rely on what is already known about the neuronal activity (as in the temporal-summation experiments discussed below). Although the model is at best an approximation of the complex interactions between neuronal activity, metabolic demand, and blood flow and oxygenation, it nonetheless serves as a useful platform on which to build. If the linear model were a satisfactory approximation, it would greatly simplify the analysis and interpretation of fMRI data. The relationship between fMRI and neuronal responses would be characterized completely and simply by the haemodynamic impulse response function (HIRF); that is, the fMRI response that results from a brief, spatially localized pulse of neuronal activity. Furthermore, one could estimate the underlying neuronal activity from fMRI measurements (Fig. 1). In fact, many fMRI studies have relied on the linear transform model, using measurements of the temporal component of the HIRF to estimate the underlying neuronal activity. However, these measurements of the HIRF would be worthless if the linear transform model were not a valid approximation. To be useful, the linear transform model must approximate the detailed relationship between the fMRI data and the neuronal activity. That relationship depends on three main factors, as outlined below.

Figure 1: The linear transform model of fMRI responses.
figure1

Top row: hypothetical plots of average neuronal activity over time. Bottom row: corresponding functional magnetic resonance imaging (fMRI) responses. Left: hypothetical haemodynamic impulse response function (HIRF) measured as the response to a brief pulse of neuronal activity. Right: the fMRI response when the average neuronal activity alternates (at specific times) between three different states (high, medium and low), with brief transients each time it switches from one state to another. Given the measured time course of neuronal activity, it is possible to compute (using convolution) the time course of the fMRI response. Most fMRI studies go the other way, to infer the underlying neuronal activity from the fMRI response. In the example shown, it is possible to estimate (using linear regression, given the HIRF) the relative amplitudes of the neuronal activity in each of the three states, along with the amplitude of the transients.

First, the relationship between fMRI and neuronal responses depends on the fMRI acquisition technique. Most fMRI experiments measure a blood oxygen level dependent (BOLD) response. BOLD fMRI provides a mixed signal that depends on blood flow, blood volume and blood oxygenation (Boxes 1 and 2). Some of the signal derives from changes within larger draining veins, some from smaller venules and capillaries, and some from extravascular sources15,16,17. Variations of the technique can be used to emphasize or de-emphasize one or another of these components. The intravascular signals are suppressed at higher magnetic field strengths because of the intrinsic inhomogeneity in the magnetic field within larger vessels. The acquisition can be modified to de-emphasize the BOLD signals from larger vessels, by suppressing signals that are associated with higher flow velocities15,18. Other fMRI techniques have been developed that separately measure different components of the haemodynamic response: perfusion-based fMRI measures blood flow19,20, compounds can be injected into the blood stream to allow the measurement of blood volume21, and diffusion-based fMRI techniques promise to measure changes in neural and glial cell swelling that occur with excitation22. So far, there has been only a modest amount of work towards quantifying the relationships between these different fMRI techniques (for example, see Refs 20,23,24).

Second, the relationship between fMRI and neuronal responses depends on the behavioural and stimulation protocols, and on the fMRI data-analysis methods. In the early days of fMRI there was some concern that the signal was derived entirely from large draining veins, and consequently that it would provide misleading information about the spatial localization of neuronal activity. For example, presenting visual stimuli at two nearby locations in the visual field evokes neuronal activity at two nearby locations in retinotopic visual areas such as V1. If the fMRI signal were evident only in the large vessel(s), which drain the blood from V1, then it would not be possible to distinguish between these two nearby foci of neuronal activity. Rather, the activity evoked by both visual stimuli would seem to be displaced to the location of the draining vein(s). To a large degree, this issue was resolved by adopting appropriate experimental protocols. Retinotopic maps are now routinely measured in the visual cortex using stimuli that move through the visual field, thereby evoking a travelling wave of neuronal activity across the folded grey matter surface25. Signals in a large draining vein, one that drains from the region of grey matter that corresponds to a large fraction of the retinotopic map, are de-emphasized because the blood flow and oxygenation in those vessels are roughly constant throughout the experiment. By using this travelling-wave technique, and by restricting the data analysis to the grey matter, it is possible to distinguish activity that is separated by less than 1.5 mm in the cortex25 (fMRI techniques have advanced considerably since these initial measurements were reported in 1994, allowing even finer spatial resolution, as discussed below). Hence, the experimental protocol might be crucial for obtaining precise spatial localization of functional activity, a point that we will return to later. There are surely a number of further issues concerning the experimental protocol and data analysis that can affect the fidelity of the fMRI measurements. It is therefore important, each time a new protocol is developed, to measure the relationship between the fMRI signals and the underlying neuronal activity.

Third, the relationship between fMRI and neuronal responses depends on how the neuronal activity itself is measured and quantified. Consider recording the simultaneous activity of a large number of individual neurons within a region of cortex (several millimetres) and over a period of time (several seconds). What component of the neuronal activity would be most predictive of the fMRI signal? The various possible measurements of neuronal activity include: the average firing rate of all of the neurons; the average firing rate of a subpopulation of neurons; synchronous spiking activity across the neuronal population; the local field potential (LFP); the current source density; some measure of local average synaptic activity; or some measure of the subthreshold electrical activity. What is the relationship between these different measures of neuronal activity, and how are they each related to the fMRI signal? New experiments are just beginning to shed light on this question (Box 3).

Temporal summation of fMRI responses

In a typical experiment, a time-series of BOLD fMRI images are collected while a stimulus or cognitive task is systematically varied. If the stimulus or task variations evoke a large enough change in blood flow and oxygenation in a certain brain region, then the image intensity in that region will modulate (by as much as ±5%, but typically by less than this) over time about its mean intensity value.

According to the linear transform model, it should be possible to predict the response to a long stimulus presentation by summing the responses to shorter stimuli (Fig. 2). For example, the response to a 12-s stimulus is predicted by summing the response to a 6-s stimulus with a copy of the same response delayed by 6 s. Note that this allows the possibility of an arbitrarily complex, nonlinear relationship between the stimulus and the neuronal activity. Whatever the neuronal activity is for the first 6 s, as long as it is identical for the second 6 s, it should be possible to predict the fMRI response to a pair of back-to-back stimuli, given the response to one stimulus presentation.

Figure 2: Temporal summation of fMRI responses.
figure2

Left: measured functional magnetic resonance imaging (fMRI) responses from the visual cortex for 6- and 12-s stimulus presentations. Right: the procedure for measuring temporal summation, by comparing the responses to the two different stimulus durations. BOLD, blood oxygen level dependent. Modified with permission from Ref. 13 © 1996 Society for Neuroscience.

Temporal summation holds up well in some experiments, but not in others13,20,24,26,27,28,29,30,31,32,33. The failure can be traced to the fact that very short stimulus presentations evoke disproportionately large fMRI responses, compared with what is expected from the responses to long stimulus presentations. One possible explanation for the failure of temporal summation is that although blood flow might indeed be proportional to the underlying neuronal activity, BOLD fMRI signals might have an inherently nonlinear dependence on flow (see Box 2 and Refs 20,34). In particular, the BOLD effect might saturate at high levels of blood flow because further increases in flow would cause negligible decreases in the concentration of deoxyhaemoglobin20,35,36. Consequently, a moderately strong stimulus might evoke a near-maximal fMRI response, leaving very little room to reveal any further change in fMRI signal in response to a stronger stimulus that is known to evoke more neuronal activity. Indeed, there is some evidence that flow might show temporal summation even when BOLD fMRI fails to do so20.

An alternative possibility is a failure in the design of the experiments, rather than a failure of temporal summation itself. These experiments were all completed in primary sensory and motor cortical areas, mostly in V1, where short-duration stimuli are expected to give disproportionately large neuronal responses for at least three reasons. First, V1 neurons show large transient responses after stimulus onset, which rapidly decrease over a period of several hundred milliseconds37. Second, V1 neurons adapt; their responses decrease gradually over time after prolonged (4–30 s) stimulation38. Third, V1 responses are boosted by attention (see above), and stimulus onsets are likely to engage attention automatically. Hence, very short stimulus presentations evoke disproportionately large neural responses, which, according to the linear transform model, should in turn evoke disproportionately large fMRI responses. Some studies have included a post-hoc analysis to account for the expected response transients13,20,33, but experiments have not yet been carried out that measure the temporal summation of the fMRI responses while explicitly controlling or compensating for response transients, adaptation and attention. So, the temporal-summation experiments completed so far are inconclusive because the reported failures of the linear transform model might actually be failures of the assumptions underlying the test, not failures of temporal summation itself.

Comparison of fMRI and neuronal responses

A second implication of the linear transform model is that the time course and amplitude of the fMRI responses should be predictable from the underlying neuronal responses.

Two studies have compared human fMRI and monkey single-unit data to infer that fMRI signals are directly proportional to average neuronal firing rates39,40. The first of these studies compared single-unit data recorded in monkey MT, an extrastriate area of visual cortex that is specialized for visual motion processing, with fMRI measurements in human MT+ (the MT complex, also known as V5), a motion-responsive cortical region that is widely believed to be homologous with monkey MT along with adjacent motion-sensitive areas41,42. MT neurons are sensitive to the strength, or coherence, of the motion signal. In a stimulus that consists of a field of moving dots, motion coherence can be varied parametrically from 0% (in which each dot moves in a random direction) to 100% (in which all dots move in the same direction). The responses of MT neurons to such stochastic motion stimuli have been extensively characterized43. Using this large database of single-unit recordings, Rees et al.39 calculated that the average neuronal firing rate across a large population of MT neurons would be expected to increase linearly with coherence. They then confirmed that the fMRI responses in human MT+ also increased approximately linearly with motion coherence, indicating a proportionality between fMRI signals and average firing rates.

Motivated by their report, we carried out a similar analysis of some of our previously reported measurements of the activity in human and monkey primary visual cortex (V1)40,44,45. We used stimulus contrast as the independent variable, because nearly all V1 neurons (along with most neurons in other early visual areas) increase their firing rates monotonically with contrast. We would expect, therefore, that if the fMRI signals were proportional to average firing rates, then they too would increase monotonically with stimulus contrast. We analysed the responses of a large database of macaque V1 neurons to stimuli of varying contrasts, orientations, directions and spatiotemporal frequencies. From this database of individual V1 neurons, we estimated the average firing rate in monkey V1, as a function of contrast, for the specific stimulus pattern that had been used in the fMRI experiments. The fMRI and average firing-rate response curves superimposed, again indicating a proportional relationship between fMRI response and average firing rate. However, both of these studies relied on a series of computations to estimate the average neuronal activity from the electrophysiological data. Moreover, they compared data not only from different areas of cortical tissue, but also from different brains in two different species.

Several studies have found generally good agreement between the amplitudes of fMRI and electroencephalographic (EEG) signals, in both humans and anaesthetized rats27,46. Likewise, blood flow measured with the laser Doppler flow technique has been reported to be highly predictable from EEG and LFP measurements47,48,49. See Box 3 for a discussion of how the EEG and LFP measurements relate to the spiking activity of neurons.

A recent, seminal study by Logothetis et al.50 evaluated the linear transform model by comparing simultaneously recorded fMRI and neuronal signals in the anaesthetized monkey. Neuronal activity was recorded with a microelectrode in V1, and analysed to measure multi-unit activity (MUA) and LFP, which were extracted from different frequency bands of the raw electrophysiological signal (LFP, 40–130 Hz; MUA, 300–1,500 Hz). For each recording site, the fMRI data were analysed over a small (approximately 2 × 2.5 × 7 mm) region of grey matter surrounding the electrode tip.

Logothetis et al. examined how well they could predict the fMRI measurements from the LFP and MUA measurements (Fig. 3). They measured baseline activity in the absence of visual stimulation at each recording site to estimate the HIRFs separately for the MUA and LFP (the method they used to estimate the HIRFs relied on having simultaneous fMRI and electrophysiological measurements). They then used separate measurements of visually evoked activity to evaluate the linear transform model, by comparing the visually evoked fMRI responses with those predicted from the MUA and LFP signals (according to the linear transform model, given the estimated HIRFs). At some recording sites, the linear transform model predicted the measured fMRI responses well, explaining more than 90% of the variance in the fMRI signals. However, at other recording sites, the LFP and MUA signals were very poor predictors of the fMRI responses. At those sites where the LFP was a good predictor of the fMRI responses, MUA also tended to be a good predictor (N. K. Logothetis, personal communication). On average, the LFP was slightly better than MUA at predicting the fMRI responses (although this difference was statistically significant, the LFP accounted for only 7.6% more of the variance in the fMRI responses than did the MUA). The data indicate that this difference related to the transient nature of the MUA. Around one-quarter of the recording sites showed transient MUA responses that returned to baseline within 2–4 s even though the stimulus stayed on for much longer. The LFP and fMRI signals from the same sites were both sustained throughout the stimulus presentations, which lasted for up to 20 s.

Figure 3: Neuronal versus fMRI responses.
figure3

a | Haemodynamic impulse response functions (HIRFs) estimated from two (red and blue curves) different recording sites. b | Comparison of the measured functional magnetic resonance imaging (fMRI) response with that predicted (using the linear transform model and the estimated HIRF) from the local field potential (LFP) (BOLD, blood oxygen level dependent; s.d., standard deviation). Stimulus was a full-field, rotating, checkerboard pattern that varied in contrast over time. Dark turquoise rectangles: contrast level over time. Light turquoise: measured neuronal activity (LFP). Red curve: measured fMRI response. Blue curve: predicted fMRI response. Reproduced with permission from Ref. 50 © 2001 Macmillan Magazines Ltd.

On the basis of these results, Logothetis et al. concluded that BOLD fMRI signals “reflect the input and intracortical processing of a given area rather than its spiking output.” Specifically, the data reported in most single-unit electrophysiological experiments reflect the activity of a biased sample of neurons, mostly large pyramidal cells that send projections to other brain areas. BOLD signals, they argue, also reflect the activity of incoming fibres, interneurons, and recurrent excitatory or inhibitory activity, rather than just 'output' firing rates of the pyramidal cells (N. K. Logothetis, personal communication). Under some conditions, the intracortical processing within a region of cortical tissue will be highly correlated with the output spiking activity of the pyramidal cells. For example, the inputs, intracortical activity and outputs in V1 all increase monotonically with stimulus contrast (see above). But there might be some circumstances in which there is a decoupling of inputs and intracortical processing from output activity. For example, the strong attentional signals (mentioned above) measured with fMRI in early visual cortical areas might reflect subthreshold activity arising from modulatory inputs that do not evoke spikes in the output pyramidal cells. The interpretation of the fMRI results depends crucially, therefore, on the extent to which the output from a cortical area can be decoupled from the intracortical activity within that area.

An example of such a decoupling of inputs from outputs was reported by Mathiesen et al.47, who compared blood flow, LFP and spiking activity in the cerebellum of anaesthetized rats, while electrically stimulating inputs to the cerebellum. The crucial condition involved stimulating the cerebellar parallel fibres that evoke monosynaptic excitatory postsynaptic potentials and disynaptic inhibitory postsynaptic potentials in Purkinje cells, the primary output neurons of the cerebellar cortex. By taking advantage of this specialized cerebellar circuitry, they provided strong synaptic excitation, including excitation of inhibitory interneurons, with little or no change in the net activity of the Purkinje cells. Nevertheless, blood flow and LFP increased.

Perhaps the most important result in the Logothetis et al. study concerns the large differences across recording sites. This raises a number of questions. Are the estimated HIRFs stable across recording sites? Is the fMRI signal less well predicted by neuronal activity at some sites than at others because of their proximity to larger veins? In this context, it is also important to note that the stimuli were presented in the commonly used block-alternation design, with relatively long (4–24 s) intervals of a full-field checkerboard pattern that alternated with intervals of a blank (uniform grey) field. The results might differ if a rapid series of brief (1 s) stimulus presentations is used (N. K. Logothetis, personal communication). Under such conditions the transient activity would dominate, thereby increasing the correlation between the LFP, MUA and single-unit activity, and causing each of these measures to be more predictive of the fMRI responses. This would be welcome news because a growing number of fMRI experiments have been using such a rapid sequence of stimulus presentations with brief inter-trial intervals30,51.

A final result from the Logothetis et al. study is that fMRI and neuronal responses were not simply proportional to one another. A 12% stimulus contrast evoked about half the maximum fMRI response but much less than half the maximum LFP and MUA. This implies that there is a monotonic but nonlinear relationship between the fMRI signal and neuronal responses, such that the fMRI responses increase more rapidly than neuronal responses at low contrasts, but less rapidly at high contrasts. For example, the fMRI signal might include a contribution of asynchronous, subthreshold activity that is not visible in LFP or MUA measurements. However, just the opposite result was observed in a comparison of LFP and blood flow (measured with the laser Doppler flow technique) in the somatosensory cortex of the anaesthetized rat. No blood flow response was elicited at low levels of neuronal activity, whereas flow increased in proportion to LFP above a certain threshold level49. This indicates that flow increases more slowly than neuronal activity at low levels of activity, but that flow increases at a greater rate for higher levels of activity.

Despite the nonlinear relationships observed between LFP and either flow or BOLD signals, single-unit firing rates seem to be roughly proportional to BOLD fMRI signals. Isolated V1 neurons reach half their maximum firing rate at 10–15% contrast on average40,45, consistent with the fMRI measurements reported by Logothetis et al. and with other fMRI measurements in humans40,44, but in contrast to the LFP and MUA measurements reported by Logothetis et al.

Colocalization of fMRI and neuronal activity

A third implication of the linear transform model is that the fMRI responses should be colocalized with the underlying neuronal activity. This has been explored in both human and monkey brains.

Several studies of patients undergoing neurosurgery have reported general agreement between sensorimotor maps measured with fMRI and intraoperative cortical-surface recording, cortical stimulation and optical imaging52,53,54,55,56. This has obvious and important implications for pre-surgical planning. However, Cannestra et al.56 point out that the precision of the concordance depends crucially on the fMRI data-analysis techniques. As is typical of many fMRI studies, they used a model of the HIRF to analyse their data and localize regions of activity. Using one model of the HIRF, the CENTROID of fMRI activity was on average more than 1 cm from the centroid of intrinsic optical-imaging activity. A slight modification to the HIRF, including an initial dip (Box 2), yielded a marked improvement such that the centroids were separated by 2 mm on average.

Disbrow et al.57 reported a profound discordance between fMRI and electrophysiological recordings of somatosensory cortical maps in the anaesthetized monkey. Although it was possible to use fMRI to determine the gross organization of body-part representations, there were important differences between the two maps. First, in only about half of the measurements were the centroids of fMRI activity located within the respective hand and face representations obtained electrophysiologically, and the mean distance between map centres (fMRI versus electrophysiology) was 1 cm. Second, the fMRI-defined cortical maps were considerably larger than the maps that were obtained on the basis of microelectrode recordings.

Wandell et al.58, on the other hand, reported excellent agreement between the locations and retinotopic maps of monkey V1, V2, V4 and MT, as measured by fMRI, and those expected from the electrophysiological literature. This difference between the accuracy of the somatosensory and visual maps might reflect a fundamental difference in the basis of the haemodynamic responses in these two cortical areas. Or it might be a simple consequence of differences in the experimental methods; Wandell et al. used the travelling-wave stimulus to measure the retinotopic maps in visual cortex, whereas as Disbrow et al. presented tactile stimuli using a block-alternation design with relatively long (20 s) intervals of stimulation versus rest.

Further evidence for the colocalization of fMRI and neuronal signals comes from experiments in which fMRI has been used to image ocular dominance and orientation columns59,60,61,62,63. Although some of these papers have provoked considerable controversy (for example, Refs 64,65), others are compelling and would not have been possible unless the signals were properly localized to within a fraction of a millimetre.

Conclusions

We started this review by describing the linear transform model, which has been a common assumption in the interpretation of fMRI studies. Does the evidence indicate that this model is a reasonable and useful approximation? The answer is 'yes' for some recording sites, in some brain areas, using certain experimental protocols. But the answer is 'no' under other circumstances.

After ten years of fMRI studies, there is still much to learn about the source of these signals. In particular, further measurements that compare fMRI and neuronal signals are needed. The specifics of the experimental protocols are important and, hence, should be systematically varied. Clearly, we need to know more about the relationships between the different measures of neuronal activity, single- and multi-unit spiking activity, and LFP, to determine whether they reflect different aspects of neuronal function. For example, it would be helpful for neurophysiologists routinely to digitize the entire electrophysiological signal, not just the spike times, to determine whether or not, and under which circumstances, the intracortical activity within a cortical area (for example, as represented by the LFP or MUA) might be decoupled from the output spiking activity of isolated single units. It would also be helpful to compare measurements of local average firing rates with local average synaptic activity, because synaptic activity has been identified as being linked with metabolic demand (Box 2), although it is not clear exactly how average synaptic activity could be measured directly. In addition, it would be informative to measure the fMRI signal in the absence of spikes — for example, by cooling the cortex — to determine how much of the signal is induced exclusively by the inputs to a cortical area.

This review has raised more questions than answers. Although these questions are motivated by methodological considerations, many of them bear on fundamental issues concerning cortical circuitry and the ways in which large populations of neurons represent information in the brain. With any new technique, it takes some time to sort out the methodological details. MRI offers a huge amount of flexibility — good news, because it should allow us to refine the methods to measure precisely the underlying neuronal activity, but also bad news, because it might take a while for us to converge on those methods. fMRI provides a new and different picture of brain function, complementary to other techniques; although this presents a challenge, it will also prove to be of great benefit.

References

  1. 1

    Ogawa, S., Lee, T. M., Kay, A. R. & Tank, D. W. Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proc. Natl Acad. Sci. USA 87, 9868–9872 (1990).The first demonstration of BOLD MRI. This paper was followed two years later by several concurrent reports (references 2–5 ) of the use of BOLD fMRI to measure activity non-invasively in the human brain.

    CAS  Article  PubMed  Google Scholar 

  2. 2

    Ogawa, S. et al. Intrinsic signal changes accompanying sensory stimulation: functional brain mapping with magnetic resonance imaging. Proc. Natl Acad. Sci. USA 89, 5951–5955 (1992).

    CAS  Article  PubMed  Google Scholar 

  3. 3

    Kwong, K. K. et al. Dynamic magnetic resonance imaging of human brain activity during primary sensory stimulation. Proc. Natl Acad. Sci. USA 89, 5675–5679 (1992).

    CAS  Article  PubMed  Google Scholar 

  4. 4

    Bandettini, P. A., Wong, E. C., Hinks, R. S., Tikofsky, R. S. & Hyde, J. S. Time course EPI of human brain function during task activation. Magn. Reson. Med. 25, 390–397 (1992).

    CAS  Article  PubMed  Google Scholar 

  5. 5

    Blamire, A. M. et al. Dynamic mapping of the human visual cortex by high-speed magnetic resonance imaging. Proc. Natl Acad. Sci. USA 89, 11069–11073 (1992).

    CAS  Article  PubMed  Google Scholar 

  6. 6

    Kanwisher, N. & Wojciulik, E. Visual attention: insights from brain imaging. Nature Rev. Neurosci. 1, 91–100 (2000).

    CAS  Article  Google Scholar 

  7. 7

    Luck, S. J., Chelazzi, L., Hillyard, S. A. & Desimone, R. Neural mechanisms of spatial selective attention in areas V1, V2, and V4 of macaque visual cortex. J. Neurophysiol. 77, 24–42 (1997).

    CAS  Article  PubMed  Google Scholar 

  8. 8

    McAdams, C. J. & Maunsell, J. H. R. Effects of attention on orientation-tuning functions of single neurons in macaque cortical area V4. J. Neurosci. 19, 431–441 (1999).

    CAS  Article  PubMed  Google Scholar 

  9. 9

    Kastner, S., Pinsk, M. A., De Weerd, P., Desimone, R. & Ungerleider, L. G. Increased activity in human visual cortex during directed attention in the absence of visual stimulation. Neuron 22, 751–761 (1999).

    CAS  Article  PubMed  Google Scholar 

  10. 10

    Ress, D., Backus, B. T. & Heeger, D. J. Activity in primary visual cortex predicts performance in a visual detection task. Nature Neurosci. 3, 940–945 (2000).

    CAS  Article  PubMed  Google Scholar 

  11. 11

    Fries, P., Reynolds, J. H., Rorie, A. E. & Desimone, R. Modulation of oscillatory neuronal synchronization by selective visual attention. Science 291, 1560–1563 (2001).

    CAS  Article  PubMed  Google Scholar 

  12. 12

    Scannell, J. W. & Young, M. P. Neuronal population activity and functional imaging. Proc. R. Soc. Lond. B 266, 875–881 (1999).

    CAS  Article  Google Scholar 

  13. 13

    Boynton, G. M., Engel, S. A., Glover, G. H. & Heeger, D. J. Linear systems analysis of functional magnetic resonance imaging in human V1. J. Neurosci. 16, 4207–4221 (1996).Framed the linear transform hypothesis and reported the first measurements of temporal summation.

    CAS  Article  PubMed  Google Scholar 

  14. 14

    Friston, K. J., Jezzard, P. & Turner, R. Analysis of functional MRI time-series. Hum. Brain Mapp. 1, 153–171 (1994).Developed methods, widely used to analyse fMRI data, that rely on the linear transform hypothesis.

    Article  Google Scholar 

  15. 15

    Boxerman, J. L. et al. The intravascular contribution to fMRI signal change: Monte Carlo modeling and diffusion-weighted studies in vivo. Magn. Reson. Med. 34, 4–10 (1995).

    CAS  Article  PubMed  Google Scholar 

  16. 16

    Zhong, J., Kennan, R. P., Fulbright, R. K. & Gore, J. C. Quantification of intravascular and extravascular contributions to BOLD effects induced by alteration in oxygenation or intravascular contrast agents. Magn. Reson. Med. 40, 526–536 (1998).

    CAS  Article  PubMed  Google Scholar 

  17. 17

    Hoogenraad, F. G. et al. Quantitative differentiation between BOLD models in fMRI. Magn. Reson. Med. 45, 233–246 (2001).

    CAS  Article  PubMed  Google Scholar 

  18. 18

    Song, A. W., Wong, E. C., Tan, S. G. & Hyde, J. S. Diffusion weighted fMRI at 1.5 T. Magn. Reson. Med. 35, 155–158 (1996).

    CAS  Article  PubMed  Google Scholar 

  19. 19

    Kim, S. G. Quantification of relative cerebral blood flow change by flow-sensitive alternating inversion recovery (FAIR) technique: application to functional mapping. Magn. Reson. Med. 34, 293–301 (1995).

    CAS  Article  PubMed  Google Scholar 

  20. 20

    Miller, K. L. et al. Nonlinear temporal dynamics of the cerebral blood flow response. Hum. Brain Mapp. 13, 1–12 (2001).

    CAS  Article  PubMed  Google Scholar 

  21. 21

    Mandeville, J. B. et al. Dynamic functional imaging of relative cerebral blood volume during rat forepaw stimulation. Magn. Reson. Med. 39, 615–624 (1998).

    CAS  Article  PubMed  Google Scholar 

  22. 22

    Darquie, A., Poline, J. B., Poupon, C., Saint-Jalmes, H. & Le Bihan, D. Transient decrease in water diffusion observed in human occipital cortex during visual stimulation. Proc. Natl Acad. Sci. USA 98, 9391–9395 (2001).

    CAS  Article  PubMed  Google Scholar 

  23. 23

    Luh, W. M., Wong, E. C., Bandettini, P. A., Ward, B. D. & Hyde, J. S. Comparison of simultaneously measured perfusion and BOLD signal increases during brain activation with T1-based tissue identification. Magn. Reson. Med. 44, 137–143 (2000).

    CAS  Article  PubMed  Google Scholar 

  24. 24

    Hoge, R. D. et al. Stimulus-dependent BOLD and perfusion dynamics in human V1. Neuroimage 9, 573–585 (1999).

    CAS  Article  PubMed  Google Scholar 

  25. 25

    Engel, S. A. et al. fMRI of human visual cortex. Nature 369, 525 (1994).Introduced the travelling-wave stimulation protocol for measuring retinotopic maps in human visual cortex, and resolved the 'brain–vein' debate by showing that one could reliably distinguish activity separated by <1.5 mm in cortex.

    CAS  Article  PubMed  Google Scholar 

  26. 26

    Glover, G. H. Deconvolution of impulse response in event-related BOLD fMRI. Neuroimage 9, 416–429 (1999).

    CAS  Article  PubMed  Google Scholar 

  27. 27

    Ogawa, S. et al. An approach to probe some neural systems interaction by functional MRI at neural time scale down to milliseconds. Proc. Natl Acad. Sci. USA 97, 11026–11031 (2000).

    CAS  Article  PubMed  Google Scholar 

  28. 28

    Kershaw, J., Kashikura, K., Zhang, X., Abe, S. & Kanno, I. Bayesian technique for investigating linearity in event-related BOLD fMRI. Magn. Reson. Med. 45, 1081–1094 (2001).

    CAS  Article  PubMed  Google Scholar 

  29. 29

    Friston, K. J., Josephs, O., Rees, G. & Turner, R. Nonlinear event-related responses in fMRI. Magn. Reson. Med. 39, 41–52 (1998).

    CAS  Article  PubMed  Google Scholar 

  30. 30

    Dale, A. M. & Buckner, R. L. Selective averaging of rapidly presented individual trials using fMRI. Hum. Brain Mapp. 5, 329–340 (1997).

    CAS  Article  PubMed  Google Scholar 

  31. 31

    Robson, M. D., Dorosz, J. L. & Gore, J. C. Measurements of the temporal fMRI response of the human auditory cortex to trains of tones. Neuroimage 7, 185–198 (1998).

    CAS  Article  PubMed  Google Scholar 

  32. 32

    Vazquez, A. L. & Noll, D. C. Nonlinear aspects of the BOLD response in functional MRI. Neuroimage 7, 108–118 (1998).

    CAS  Article  PubMed  Google Scholar 

  33. 33

    Birn, R. M., Saad, Z. S. & Bandettini, P. A. Spatial heterogeneity of the nonlinear dynamics in the fMRI bold response. Neuroimage 14, 817–826 (2001).

    CAS  Article  PubMed  Google Scholar 

  34. 34

    Mechelli, A., Price, C. J. & Friston, K. J. Nonlinear coupling between evoked rCBF and BOLD signals: a simulation study of hemodynamic responses. Neuroimage 14, 862–872 (2001).

    CAS  Article  PubMed  Google Scholar 

  35. 35

    Rees, G. et al. Characterizing the relationship between BOLD contrast and regional cerebral blood flow measurements by varying the stimulus presentation rate. Neuroimage 6, 270–278 (1997).

    CAS  Article  PubMed  Google Scholar 

  36. 36

    Davis, T. L., Kwong, K. K., Weisskoff, R. M. & Rosen, B. R. Calibrated functional MRI: mapping the dynamics of oxidative metabolism. Proc. Natl Acad. Sci. USA 95, 1834–1839 (1998).

    CAS  Article  PubMed  Google Scholar 

  37. 37

    Muller, J. R., Metha, A. B., Krauskopf, J. & Lennie, P. Information conveyed by onset transients in responses of striate cortical neurons. J. Neurosci. 21, 6978–6990 (2001).

    CAS  Article  PubMed  Google Scholar 

  38. 38

    Ohzawa, I., Sclar, G. & Freeman, R. D. Contrast gain control in the cat visual cortex. Nature 298, 266–268 (1982).

    CAS  Article  PubMed  Google Scholar 

  39. 39

    Rees, G., Friston, K. & Koch, C. A direct quantitative relationship between the functional properties of human and macaque V5. Nature Neurosci. 3, 716–723 (2000).

    CAS  Article  PubMed  Google Scholar 

  40. 40

    Heeger, D. J., Huk, A. C., Geisler, W. S. & Albrecht, D. G. Spikes versus BOLD: what does neuroimaging tell us about neuronal activity? Nature Neurosci. 3, 631–633 (2000).

    CAS  Article  PubMed  Google Scholar 

  41. 41

    Zeki, S. et al. A direct demonstration of functional specialization in human visual cortex. J. Neurosci. 11, 641–649 (1991).

    CAS  Article  PubMed  Google Scholar 

  42. 42

    Huk, A. C., Ress, D. & Heeger, D. J. Neuronal basis of the motion aftereffect reconsidered. Neuron 32, 161–172 (2001).

    CAS  Article  PubMed  Google Scholar 

  43. 43

    Britten, K. H. & Newsome, W. T. Tuning bandwidths for near-threshold stimuli in area MT. J. Neurophysiol. 80, 762–770 (1998).

    CAS  Article  PubMed  Google Scholar 

  44. 44

    Boynton, G. M., Demb, J. B., Glover, G. H. & Heeger, D. J. Neuronal basis of contrast discrimination. Vision Res. 39, 257–269 (1999).

    CAS  Article  PubMed  Google Scholar 

  45. 45

    Geisler, W. S. & Albrecht, D. G. Visual cortex neurons in monkeys and cats: detection, discrimination, and identification. Vis. Neurosci. 14, 897–919 (1997).

    CAS  Article  PubMed  Google Scholar 

  46. 46

    Brinker, G. et al. Simultaneous recording of evoked potentials and T2*-weighted MR images during somatosensory stimulation of rat. Magn. Reson. Med. 41, 469–473 (1999).

    CAS  Article  PubMed  Google Scholar 

  47. 47

    Mathiesen, C., Caesar, K., Akgoren, N. & Lauritzen, M. Modification of activity-dependent increases of cerebral blood flow by excitatory synaptic activity and spikes in rat cerebellar cortex. J. Physiol. (Lond.) 512, 555–566 (1998).A compelling demonstration that blood flow can be decoupled from spiking activity.

    CAS  Article  Google Scholar 

  48. 48

    Ances, B. M., Zarahn, E., Greenberg, J. H. & Detre, J. A. Coupling of neural activation to blood flow in the somatosensory cortex of rats is time-intensity separable, but not linear. J. Cereb. Blood Flow Metab. 20, 921–930 (2000).

    CAS  Article  PubMed  Google Scholar 

  49. 49

    Nielsen, A. N. & Lauritzen, M. Coupling and uncoupling of activity-dependent increases of neuronal activity and blood flow in rat somatosensory cortex. J. Physiol. (Lond.) 553, 773–785 (2001). | PubMed |

    Article  Google Scholar 

  50. 50

    Logothetis, N. K., Pauls, J., Augath, M., Trinath, T. & Oeltermann, A. Neurophysiological investigation of the basis of the fMRI signal. Nature 412, 150–157 (2001).Evaluated the linear transform model by comparing simultaneously recorded fMRI and neuronal signals in the anaesthetized monkey.

    CAS  Article  PubMed  Google Scholar 

  51. 51

    Burock, M. A., Buckner, R. L., Woldorff, M. G., Rosen, B. R. & Dale, A. M. Randomized event-related experimental designs allow for extremely rapid presentation rates using functional MRI. Neuroreport 9, 3735–3739 (1998).

    CAS  Article  PubMed  Google Scholar 

  52. 52

    Puce, A. et al. Functional magnetic resonance imaging of sensory and motor cortex: comparison with electrophysiological localization. J. Neurosurg. 83, 262–270 (1995).

    CAS  Article  PubMed  Google Scholar 

  53. 53

    Yousry, T. A. et al. Topography of the cortical motor hand area: prospective study with functional MR imaging and direct motor mapping at surgery. Radiology 195, 23–29 (1995).

    CAS  Article  PubMed  Google Scholar 

  54. 54

    Schulder, M. et al. Functional image-guided surgery of intracranial tumors located in or near the sensorimotor cortex. J. Neurosurg. 89, 412–418 (1998).

    CAS  Article  PubMed  Google Scholar 

  55. 55

    Ruge, M. I. et al. Concordance between functional magnetic resonance imaging and intraoperative language mapping. Stereotact. Funct. Neurosurg. 72, 95–102 (1999).

    CAS  Article  PubMed  Google Scholar 

  56. 56

    Cannestra, A. F. et al. Temporal spatial differences observed by functional MRI and human intraoperative optical imaging. Cereb. Cortex 11, 773–782 (2001).

    CAS  Article  PubMed  Google Scholar 

  57. 57

    Disbrow, E. A., Slutsky, D. A., Roberts, T. P. & Krubitzer, L. A. Functional MRI at 1.5 Tesla: a comparison of the blood oxygenation level-dependent signal and electrophysiology. Proc. Natl Acad. Sci. USA 97, 9718–9723 (2000).Reported a profound discordance between fMRI and electrophysiological recordings of somatosensory cortical maps in the anaesthetized monkey.

    CAS  Article  PubMed  Google Scholar 

  58. 58

    Wandell, B., Press, W., Brewer, A. & Logothetis, N. fMRI measurements of visual areas and retinotopic maps in monkey. Soc. Neurosci. Abstr. 30, 309.9 (2000).

  59. 59

    Cheng, K., Waggoner, R. A. & Tanaka, K. Human ocular dominance columns as revealed by high-field functional magnetic resonance imaging. Neuron 32, 359–374 (2001).The most compelling demonstration so far of the use of fMRI to image human ocular dominance.

    CAS  Article  PubMed  Google Scholar 

  60. 60

    Menon, R. S., Ogawa, S., Strupp, J. P. & Ugurbil, K. Ocular dominance in human V1 demonstrated by functional magnetic resonance imaging. J. Neurophysiol. 77, 2780–2787 (1997).

    CAS  Article  PubMed  Google Scholar 

  61. 61

    Goodyear, B. G. & Menon, R. S. Brief visual stimulation allows mapping of ocular dominance in visual cortex using fMRI. Hum. Brain Mapp. 14, 210–217 (2001).

    CAS  Article  PubMed  Google Scholar 

  62. 62

    Menon, R. S. & Goodyear, B. G. Submillimeter functional localization in human striate cortex using BOLD contrast at 4 Tesla: implications for the vascular point-spread function. Magn. Reson. Med. 41, 230–235 (1999).

    CAS  Article  PubMed  Google Scholar 

  63. 63

    Kim, D. S., Duong, T. Q. & Kim, S. G. High-resolution mapping of iso-orientation columns by fMRI. Nature Neurosci. 3, 164–169 (2000).

    CAS  Article  PubMed  Google Scholar 

  64. 64

    Logothetis, N. Can current fMRI techniques reveal the micro-architecture of cortex? Nature Neurosci. 3, 413–414 (2000).

    CAS  Article  PubMed  Google Scholar 

  65. 65

    Kim, D. S., Duong, T. Q. & Kim, S. G. Reply to: Can current fMRI techniques reveal the micro-architecture of cortex? Nature Neurosci. 3, 414 (2000). | PubMed |

  66. 66

    Malonek, D. & Grinvald, A. Interactions between electrical activity and cortical microcirculation revealed by imaging spectroscopy: implications for functional brain mapping. Science 272, 551–554 (1996).

    CAS  Article  PubMed  Google Scholar 

  67. 67

    Malonek, D. et al. Vascular imprints of neuronal activity: relationships between the dynamics of cortical blood flow, oxygenation, and volume changes following sensory stimulation. Proc. Natl Acad. Sci. USA 94, 14826–14831 (1997).

    CAS  Article  PubMed  Google Scholar 

  68. 68

    Vanzetta, I. & Grinvald, A. Increased cortical oxidative metabolism due to sensory stimulation: implications for functional brain imaging. Science 286, 1555–1558 (1999).The most compelling in a series of influential papers about the initial dip, the proposed initial increase in oxygen consumption that occurs immediately after the onset of neuronal activity. See also references 66, 67 and 87.

    CAS  Article  PubMed  Google Scholar 

  69. 69

    Fox, P. T. & Raichle, M. E. Focal physiological uncoupling of cerebral blood flow and oxidative metabolism during somatosensory stimulation in human subjects. Proc. Natl Acad. Sci. USA 83, 1140–1144 (1986).

    CAS  Article  PubMed  Google Scholar 

  70. 70

    Fox, P. T., Raichle, M. E., Mintun, M. A. & Dence, C. Nonoxidative glucose consumption during focal physiologic neural activity. Science 241, 462–464 (1988).A classic study showing that oxygen consumption did not match blood flow and glucose consumption, indicating that increases in brain activation are supported by non-oxidative glucose metabolism. See also reference 69.

    CAS  Article  PubMed  Google Scholar 

  71. 71

    Buxton, R. B., Wong, E. C. & Frank, L. R. Dynamics of blood flow and oxygenation changes during brain activation: the balloon model. Magn. Reson. Med. 39, 855–864 (1998).An influential model of blood flow, volume and oxygenation that explains many of the properties of the haemodynamic response. See also references 72 and 73.

    CAS  Article  PubMed  Google Scholar 

  72. 72

    Friston, K. J., Mechelli, A., Turner, R. & Price, C. J. Nonlinear responses in fMRI: the balloon model, volterra kernels, and other hemodynamics. Neuroimage 12, 466–477 (2000).

    CAS  Article  PubMed  Google Scholar 

  73. 73

    Mandeville, J. B. et al. Evidence of a cerebrovascular postarteriole windkessel with delayed compliance. J. Cereb. Blood Flow Metab. 19, 679–689 (1999).

    CAS  Article  PubMed  Google Scholar 

  74. 74

    Mandeville, J. B. et al. MRI measurement of the temporal evolution of relative CMRO2 during rat forepaw stimulation. Magn. Reson. Med. 42, 944–951 (1999).

    CAS  Article  PubMed  Google Scholar 

  75. 75

    Kim, S. G., Rostrup, E., Larsson, H. B., Ogawa, S. & Paulson, O. B. Determination of relative CMRO2 from CBF and BOLD changes: significant increase of oxygen consumption rate during visual stimulation. Magn. Reson. Med. 41, 1152–1161 (1999).

    CAS  Article  PubMed  Google Scholar 

  76. 76

    Kim, S. G. & Ugurbil, K. Comparison of blood oxygenation and cerebral blood flow effects in fMRI: estimation of relative oxygen consumption change. Magn. Reson. Med. 38, 59–65 (1997).

    CAS  Article  PubMed  Google Scholar 

  77. 77

    Kennan, R. P., Scanley, B. E., Innis, R. B. & Gore, J. C. Physiological basis for BOLD MR signal changes due to neuronal stimulation: separation of blood volume and magnetic susceptibility effects. Magn. Reson. Med. 40, 840–846 (1998).

    CAS  Article  PubMed  Google Scholar 

  78. 78

    Lee, S. P., Duong, T. Q., Yang, G., Iadecola, C. & Kim, S. G. Relative changes of cerebral arterial and venous blood volumes during increased cerebral blood flow: implications for BOLD fMRI. Magn. Reson. Med. 45, 791–800 (2001).

    CAS  Article  PubMed  Google Scholar 

  79. 79

    Bandettini, P. A. et al. Characterization of cerebral blood oxygenation and flow changes during prolonged brain activation. Hum. Brain Mapp. 5, 93–109 (1997).

    CAS  Article  PubMed  Google Scholar 

  80. 80

    Mandeville, J. B. et al. Regional sensitivity and coupling of BOLD and CBV changes during stimulation of rat brain. Magn. Reson. Med. 45, 443–447 (2001).

    CAS  Article  PubMed  Google Scholar 

  81. 81

    Marota, J. J. et al. Investigation of the early response to rat forepaw stimulation. Magn. Reson. Med. 41, 247–252 (1999).

    CAS  Article  PubMed  Google Scholar 

  82. 82

    Jones, M., Berwick, J., Johnston, D. & Mayhew, J. Concurrent optical imaging spectroscopy and laser-Doppler flowmetry: the relationship between blood flow, oxygenation, and volume in rodent barrel cortex. Neuroimage 13, 1002–1015 (2001).

    CAS  Article  PubMed  Google Scholar 

  83. 83

    Hu, X., Le, T. H. & Ugurbil, K. Evaluation of the early response in fMRI in individual subjects using short stimulus duration. Magn. Reson. Med. 37, 877–884 (1997).

    CAS  Article  PubMed  Google Scholar 

  84. 84

    Logothetis, N. K., Guggenberger, H., Peled, S. & Pauls, J. Functional imaging of the monkey brain. Nature Neurosci. 2, 555–562 (1999).

    CAS  Article  PubMed  Google Scholar 

  85. 85

    Menon, R. S. et al. BOLD based functional MRI at 4 Tesla includes a capillary bed contribution: echo-planar imaging correlates with previous optical imaging using intrinsic signals. Magn. Reson. Med. 33, 453–459 (1995).

    CAS  Article  PubMed  Google Scholar 

  86. 86

    Ernst, T. & Hennig, J. Observation of a fast response in functional MR. Magn. Reson. Med. 32, 146–149 (1994).

    CAS  Article  PubMed  Google Scholar 

  87. 87

    Buxton, R. B. The elusive initial dip. Neuroimage 13, 953–958 (2001).

    CAS  Article  PubMed  Google Scholar 

  88. 88

    Schwartz, W. J. et al. Metabolic mapping of functional activity in the hypothalamo–neurohypophysial system of the rat. Science 205, 723–725 (1979).A classic study linking glucose consumption with synaptic activity.

    CAS  Article  PubMed  Google Scholar 

  89. 89

    Magistretti, P. J. & Pellerin, L. Cellular mechanisms of brain energy metabolism and their relevance to functional brain imaging. Phil. Trans. R. Soc. Lond. B 354, 1155–1163 (1999).An excellent review of the literature on neurotransmitter recycling and its relationship to neuroimaging measurements. | PubMed |

    CAS  Article  Google Scholar 

  90. 90

    Sibson, N. R. et al. Stoichiometric coupling of brain glucose metabolism and glutamatergic neuronal activity. Proc. Natl Acad. Sci. USA 95, 316–321 (1998).

    CAS  Article  PubMed  Google Scholar 

  91. 91

    Shulman, R. G. & Rothman, D. L. Interpreting functional imaging studies in terms of neurotransmitter cycling. Proc. Natl Acad. Sci. USA 95, 11993–11998 (1998).

    CAS  Article  PubMed  Google Scholar 

  92. 92

    Buxton, R. B. & Frank, L. R. A model for the coupling between cerebral blood flow and oxygen metabolism during neural stimulation. J. Cereb. Blood Flow Metab. 17, 64–72 (1997).A model showing that the apparent mismatch between blood flow and oxygen consumption might be required to support a change in oxygen metabolism, because extraction (through passive diffusion) of oxygen from the blood is less efficient at higher flow rates. See also references 93 and 94.

    CAS  Article  PubMed  Google Scholar 

  93. 93

    Hyder, F., Shulman, R. G. & Rothman, D. L. A model for the regulation of cerebral oxygen delivery. J. Appl. Physiol. 85, 554–564 (1998).

    CAS  Article  PubMed  Google Scholar 

  94. 94

    Vafaee, M. S. & Gjedde, A. Model of blood-brain transfer of oxygen explains nonlinear flow–metabolism coupling during stimulation of visual cortex. J. Cereb. Blood Flow Metab. 20, 747–754 (2000).

    CAS  Article  PubMed  Google Scholar 

  95. 95

    Hoge, R. D. et al. Investigation of BOLD signal dependence on cerebral blood flow and oxygen consumption: the deoxyhemoglobin dilution model. Magn. Reson. Med. 42, 849–863 (1999).

    CAS  Article  PubMed  Google Scholar 

  96. 96

    Attwell, D. & Laughlin, S. B. An energy budget for signaling in the grey matter of the brain. J. Cereb. Blood Flow Metab. 21, 1133–1145 (2001).An excellent review of energy consumption in the brain, which concludes that action potentials and postsynaptic effects of glutamate consume much of the energy (47% and 34%, respectively), with the resting potential consuming a smaller amount (13%) and glutamate recycling using only 3%.

    CAS  Article  PubMed  Google Scholar 

  97. 97

    Shulman, R. G., Hyder, F. & Rothman, D. L. Cerebral energetics and the glycogen shunt: neurochemical basis of functional imaging. Proc. Natl Acad. Sci. USA 98, 6417–6422 (2001).

    CAS  Article  PubMed  Google Scholar 

  98. 98

    Villringer, A. & Dirnagl, U. Coupling of brain activity and cerebral blood flow: basis of functional neuroimaging. Cerebrovasc. Brain Metab. Rev. 7, 240–276 (1995).

    CAS  PubMed  Google Scholar 

  99. 99

    Mintun, M. A. et al. Blood flow and oxygen delivery to human brain during functional activity: theoretical modeling and experimental data. Proc. Natl Acad. Sci. USA 98, 6859–6864 (2001).

    CAS  Article  PubMed  Google Scholar 

  100. 100

    Powers, W. J., Hirsch, I. B. & Cryer, P. E. Effect of stepped hypoglycemia on regional cerebral blood flow response to physiological brain activation. Am. J. Physiol. 270, H554–H559 (1996).

  101. 101

    Cholet, N., Seylaz, J., Lacombe, P. & Bonvento, G. Local uncoupling of the cerebrovascular and metabolic responses to somatosensory stimulation after neuronal nitric oxide synthase inhibition. J. Cereb. Blood Flow Metab. 17, 1191–1201 (1997).

    CAS  Article  PubMed  Google Scholar 

  102. 102

    Legatt, A. D., Arezzo, J. & Vaughan, H. G. Jr. Averaged multiple unit activity as an estimate of phasic changes in local neuronal activity: effects of volume-conducted potentials. J. Neurosci. Methods 2, 203–217 (1980).

    CAS  Article  PubMed  Google Scholar 

  103. 103

    Freeman, W. Mass Action in the Nervous System (Academic, New York, 1975).

  104. 104

    Mitzdorf, U. Properties of the evoked potential generators: current source-density analysis of visually evoked potentials in the cat cortex. Int. J. Neurosci. 33, 33–59 (1987).

    CAS  Article  PubMed  Google Scholar 

  105. 105

    Singer, W. Synchronization of cortical activity and its putative role in information processing and learning. Annu. Rev. Physiol. 55, 349–374 (1993).

    CAS  Article  PubMed  Google Scholar 

  106. 106

    Singer, W. Neuronal synchrony: a versatile code for the definition of relations? Neuron 24, 49–65; 111–125 (1999). | PubMed |

    CAS  Article  PubMed  Google Scholar 

  107. 107

    Shadlen, M. N. & Movshon, J. A. Synchrony unbound: a critical evaluation of the temporal binding hypothesis. Neuron 24, 67–77; 111–125 (1999). | PubMed |

    CAS  Article  PubMed  Google Scholar 

  108. 108

    Braitenberg, V. & Schuz, A. Anatomy of the Cortex (Springer, Berlin, 1991).

  109. 109

    Peters, A. & Sethares, C. Organization of pyramidal neurons in area 17 of monkey visual cortex. J. Comp. Neurol. 306, 1–23 (1991).

    CAS  Article  PubMed  Google Scholar 

  110. 110

    Latawiec, D., Martin, K. A. & Meskenaite, V. Termination of the geniculocortical projection in the striate cortex of macaque monkey: a quantitative immunoelectron microscopic study. J. Comp. Neurol. 419, 306–319 (2000).

    CAS  Article  PubMed  Google Scholar 

  111. 111

    Eckhorn, R. et al. Coherent oscillations: a mechanism of feature linking in the visual cortex? Multiple electrode and correlation analyses in the cat. Biol. Cybern. 60, 121–130 (1988).

    CAS  Article  PubMed  Google Scholar 

  112. 112

    Roelfsema, P. R., Engel, A. K., Konig, P. & Singer, W. Visuomotor integration is associated with zero time-lag synchronization among cortical areas. Nature 385, 157–161 (1997).

    CAS  Article  PubMed  Google Scholar 

  113. 113

    Bressler, S. L., Coppola, R. & Nakamura, R. Episodic multiregional cortical coherence at multiple frequencies during visual task performance. Nature 366, 153–156 (1993).

    CAS  Article  PubMed  Google Scholar 

Download references

Acknowledgements

We thank W. Newsome, G. DeAngelis, P. Fries, B. Wandell, G. Rees, R. Buxton, P. Bandettini, G. Glover, G. Boynton, M. Raichle and N. Logothetis for detailed comments on this manuscript. The authors are supported by grants from the National Eye Institute and the Human Frontier Science Program.

Author information

Affiliations

Authors

Corresponding author

Correspondence to David J. Heeger.

Related links

Related links

FURTHER INFORMATION

brain imaging: observing ongoing neural activity

magnetic resonance imaging 

MIT Encyclopedia of Cognitive Sciences

electrophysiology, electric and magnetic evoked fields

magnetic resonance imaging

Glossary

EXTRASTRIATE CORTEX

All visually responsive areas of cortex except the primary visual cortex.

CENTROID

Geometric centre.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Heeger, D., Ress, D. What does fMRI tell us about neuronal activity?. Nat Rev Neurosci 3, 142–151 (2002). https://doi.org/10.1038/nrn730

Download citation

Further reading