20: Functional Imaging of Cognition

• Functional Imaging Reflects the Metabolic Demand of Neural Activity

  • Functional Imaging Emerged from Studies of Blood Flow

  • Functional Imaging Reflects Energy Metabolism

• Functional Imaging Is Used to Probe Cognitive Processes

  • Imaging Perception with and Without Consciousness

  • Imaging Memory with and Without Consciousnes

  • Imaging Attentional Modulation of Conscious Perception

• Functional Imaging Has Limitations

• An Overall View

The ability of neuro-imaging to observe areas of the human brain that are active during cognitive processes has helped to stimulate the current interest in the biological underpinnings of cognitive functioning. Because invasive experiments cannot be done ethically on humans, research on the biological basis of cognitive function was until quite recently confined to laboratory animals and clinical studies of patients with cognitive disorders.

The development of techniques such as functional magnetic resonance imaging (fMRI) has made it possible to study human subjects, affording unprecedented views of the complexities of the intact working brain. Imaging of the living brain allows us to explore the behavioral significance of local neural circuits, such as cortical columns, as well as observe large-scale systems of interconnected brain regions concerned with specific mental processes such as seeing, hearing, feeling, moving, talking, and thinking.

Functional Imaging Emerged from Studies of Blood Flow

Functional imaging evolved out of seminal studies in the late 1940s by Seymour Kety and Carl F. Schmidt, who succeeded in measuring blood flow in the living brain. Although Charles S. Roy and Charles S. Sherrington earlier had found a relationship between blood flow and brain metabolism, Kety and Schmidt were the first to quantify cerebral blood flow noninvasively.

To accomplish this task, Kety and Schmidt measured the rate of cerebral blood flow by having subjects inhale nitrous oxide, a metabolically inert gas, and measuring its outflow concentration from the jugular vein (Box 20–1). In a series of landmark studies they evaluated how blood flow from the intact brain varied in different metabolic states, such as sleep and wakefulness, and in so doing they laid the foundations for modern functional imaging.

These early experiments measured only the total level of activity of the entire brain, however. They could not provide information about which parts of the brain were active, nor could they tell us whether some brain areas became more active while others became less active under set conditions.

A significant advance came in the 1970s with the introduction of positron emission tomography (PET) by Michel Ter-Pogossian, Michael Phelps, and Louis Sokoloff (Box 20–2). In the 1980s Marcus Raichle collaborated with Michael Posner to visualize the brain activity of subjects engaged in complex tasks of thought and language, thereby demonstrating that PET can be used to explore cognitive functioning.

A further advance in functional imaging occurred in 1990 when Seiji Ogawa and David Tank discovered that magnetic resonance imaging (MRI) can be made sensitive to changes in deoxyhemoglobin that are caused when neurons change their metabolic rates. They exploited the fact, first discovered in 1936 by Linus Pauling, that when oxyhemoglobin is converted to deoxyhemoglobin (by stripping hemoglobin of its four oxygen molecules), it becomes paramagnetic.

In particular, they showed that in MRI images of the hippocampal formation of anesthetized rodents areas of increased vascularity appeared darker than areas with less vascularity. When rodents breathed 100% oxygen, the image intensities in the hippocampus were brighter, thereby suggesting that these differences in intensity were caused by changes in blood oxygenation. Finally, Ogawa went on to link these differences in image intensity with metabolism. He found that systematic pharmacological alteration of basal brain metabolism in anesthetized animals induced a corresponding increase in the image intensity (Box 20–3).

In brain regions with increased metabolism the flow of oxygenated blood is greater than the consumption of oxygen, and thus leads to a relative decrease in deoxyhemoglobin. MRI areas with increased metabolism and flow of oxygenated blood appear brighter than regions that are not experiencing increased metabolism (Box 20–3). This form of functional MRI has been termed BOLD (blood oxygen level dependent) imaging.

Functional Imaging Reflects Energy Metabolism

As these discussions make clear, functional imaging does not measure neural activity but rather reflects energy metabolism, best defined as the rate at which mitochondria produce adenosine triphosphate (ATP). Because direct imaging of ATP production is difficult, functional imaging assesses correlates of energy metabolism that can be visualized with clinical imaging devices (Figure 20–1).

A surprisingly large amount of a neuron's total energy metabolism, approximately one-half, is devoted simply to maintaining the resting membrane potential—the electric potential across the cell membrane. Therefore any shift in the membrane potential will affect the rate of energy metabolism and influence functional imaging measures. The membrane potential changes when a cell fires an action potential and also in response to subthreshold excitatory or inhibitory synaptic potentials.

The remaining half of a neuron's energy metabolism is devoted to other biochemical processes, and alterations in these pathways also affect functional imaging measures, although typically on a slower time scale. These biochemical processes include all of the molecular reactions required for normal synaptic function: vesicle recycling, recruitment of second-messenger cascades, local protein synthesis, axonal transport, and transmitter release. Thus functional imaging in principle can measure the transient effect an external stimulus has on the electrical activity of neurons, as well as the more permanent effect of a disease process on neuronal biochemistry.

By visualizing the brain at work, functional imaging has transformed cognitive neuroscience. We will illustrate some insights derived from functional imaging by considering one of the field's ultimate questions—the nature of consciousness.

The idea is quite simple: By comparing brain activity between conscious and unconscious states we should be able to identify brain regions in which the activity is correlated with consciousness. Because systematically manipulating consciousness is not trivial, translating this logic into a scientific experiment is difficult. To accomplish this task, scientists have relied on the fact that exposure to the identical external stimulus can alternately evoke a conscious or unconscious experience depending on other controllable factors. For example, as you read this chapter you have blinked numerous times; nevertheless, although your brain has recorded the flickered light caused by blinking, your consciousness has not. Now, once brought to your attention, you become aware of the perceptual effects of blinking (in fact, it is now difficult for you to suppress this awareness).

Just as sensory stimuli can be processed with and without conscious perception by the brain, the recall of objects from memory can also be conscious or unconscious. Consider running into someone you met once before. Viewing the person's face may activate conscious recall of the initial meeting—the place, the time, the person's name. Or, as is often the case, you might sense that the face is familiar, but you cannot quite connect it to a time or place—the face simply does not evoke conscious recall of the initial meeting. Even worse, you might not consciously recognize the face, even though (as can be demonstrated) some regions of your brain are responding unconsciously to the face (as if your brain remembers, but you do not).

By comparing the hemodynamic response associated with perception and recall, both with and without consciousness, functional imaging studies have begun to identify regions of the brain correlated with consciousness.

Imaging Perception with and Without Consciousness

Mapping of brain function began in the middle of the 19th century, fully 100 years before the advent of functional imaging of the brain. By correlating cognitive performance with the anatomical location of brain lesions, neurologists identified brain regions involved in specific cognitive functions (see Chapter 1). However, this approach has a number of important limitations that prevented many questions about function from being answered.

For example, just because an area of primary sensory cortex may be necessary for conscious perception, because it is involved in the initial processing of sensory information, does not mean it is responsible for the conscious experience. It may simply relay sensory information to higher-order cortex that is responsible for consciousness. In principle, functional imaging can help make these distinctions.

Indeed, neural correlates of conscious perception can be measured experimentally using visual illusions in which the percept is dissociated from the physical stimulus. One such illusion results from binocular rivalry, which occurs when different visual stimuli are presented simultaneously to each eye. Typically, awareness of one or the other stimulus is suppressed so that we are consciously aware of one stimulus at a time, never both. Thus one eye's view dominates consciousness for several seconds, only to be replaced by the other eye's view. What makes binocular rivalry so remarkable is that the perceptual experience fluctuates while the physical stimulus remains constant. Because of this dissociation, binocular rivalry presents a unique opportunity for studying the neural correlates of consciousness.

What systems are recruited when one eye's view becomes dominant? According to one idea, neurons in the early stages in visual processing respond to the physical stimulus of each eye, but in later stages the signals from these neurons are switched on and off, causing the perceptual alternations. That is, a later stage serves as a "gate" to visual consciousness.

Does such a gate exist? If so, what neurons in the brain have this gating function? Are the neurons localized in particular brain areas? Are they a particular cell type? Does the gating occur through modulation of the cells' firing rates or some other component of their responses (eg, spike timing, synchronous firing)? What are the neural circuits and neural computations that support the competition between the two stimuli?

Although we do not yet have firm answers to these questions, the evoked metabolic activity of the brain under conditions of binocular rivalry has been measured with fMRI. One fMRI experiment capitalized on an interesting aspect of this perceptual phenomenon; during an alternation one typically perceives a traveling wave in which one pattern emerges initially at one location and expands progressively as it renders the other pattern invisible. The physical stimulus does not change while this conscious perceptual change is taking place—it is all "in your brain." This experiment established that waves of activity in primary visual cortex (V1) accompanied the perceptual changes during binocular rivalry.

Because the primary visual cortex is topographically organized—adjacent neurons respond to adjacent locations in the visual field (see Chapter 27)—it was possible to show that neural activity propagated over subregions of primary visual cortex. The sequential activation of these subregions correlated with the dynamic perceptual changes experienced during binocular rivalry (Figure 20–6). Similar waves of activity propagated over the immediately adjacent secondary visual areas (V2 and V3).

Another experiment showed, however, that the activity waves in primary visual cortex are not themselves sufficient for conscious perception. In this experiment subjects were temporarily distracted (their attention was diverted) so that they did not perceive the rival stimulus patterns presented to the two eyes. Waves of activity were still clearly evident in primary visual cortex, even though subjects did not experience a corresponding traveling wave. However, waves of activity were not evident in V2 and V3.

Indeed, activity in a number of brain areas other than the primary visual cortex correlates with the perceptual alternations of binocular rivalry, including higher-order visual areas in the inferior temporal lobe and areas in parietal and prefrontal cortex. It is likely that these different higher-order cortical areas play distinctive roles in visual perception during binocular rivalry. Attention, mediated by feedback from areas in parietal and prefrontal cortex, is thought to play a crucial role in coordinating the activity across these brain areas to yield conscious perceptual states, as discussed in a later section.

Perceptions are typically made up of multiple sensations, not just a single sensation isolated in experimental conditions. An introduction to a person, for example, involves visual, auditory, and often somatosensory and olfactory information, so that the conscious experience likely reflects activity in several higher-order sensory cortices. Although a multimodal percept arises from activity in numerous brain regions, we sense the conscious experience as a unified, seamless whole. The linkage between discrete functional systems in the brain that gives rise to a unified experience of consciousness is sometimes called the "binding problem." According to one view, a conscious experience occurs when neural activity in disparate regions of the brain is time locked: The activity in these areas becomes temporarily synchronous.

Imaging Memory with and Without Consciousness

Conscious perception and conscious memories have long been linked. According to one view, conscious recall occurs when a stimulus reactivates the brain regions that first encoded the conscious percept being remembered. An fMRI experiment by Randy Buckner and colleagues provided the first empirical evidence in support of this idea.

Buckner trained subjects to associate pictures or sounds with a written word. For example, the word "dog" was associated either with a picture of a dog or the sound of a dog barking. Subjects were scanned with fMRI after they were shown the word and asked to recall the associated picture or sound, thereby mapping memory-storage regions. In addition, they were scanned during exposure to the picture or sound, thereby mapping regions involved in perception.

Remarkably, first hearing and later recalling a sound stimulated some of the same higher-order regions in the auditory cortex, and viewing and later recalling a picture stimulated some of the same higher-order regions in the visual cortex. However, conscious recall of sounds or pictures did not activate areas of primary sensory cortex. These results provide evidence that conscious memory recall used some of the same regions of the brain that were used for conscious perception.

The notion that higher-order sensory cortex is recruited for memory is reinforced by studies of different stages of sleep. Although dreams are not faithful recollections of the external world, they are remarkably vivid, comparable to conscious memory. Functional imaging studies of rapid-eye-movement sleep, during which dreaming occurs, and slow-wave sleep, characterized by the absence of dreams, have found that dreaming is associated with activity throughout higher-order sensory cortical areas. Just as in Buckner's experiment, primary sensory cortex is not activated during dreaming.

Other imaging studies have found that when a stimulus induces only a sense of familiarity, not a full-blown recollection, brain activity tends to be confined to specific sensory regions representing one, or at most, a very few modalities. Taken together these studies demonstrate that, just as with conscious perception, simultaneous activity in several higher-order sensory regions underlies recollection (conscious recall of the stimulus along with the associated details of the context, when and where the stimulus was initially perceived, what else happened at the same time, etc.).

Imaging Attentional Modulation of Conscious Perception

Our brain is constantly bombarded by external and internal stimulation, yet at any given moment we are only aware of a small fraction of this input. Attention is one factor that influences the focus and scope of our awareness. As we mentioned earlier in Chapter 17, the American psychologist William James defined attention as "… the taking possession by the mind, in clear and vivid form, of one out of what seem several simultaneous possible objects or trains of thought." James captured the key element of attention—when confronted with more than one input the brain does not process all inputs equally.

Performance on a wide variety of perceptual discrimination and identification tasks is faster and more accurate when subjects attend to the right place at the right time. Several investigators have developed experimental protocols to characterize the behavioral consequences of attention. For example, in a visual attention experiment the subject is asked to fixate a spot at the center of a computer screen while visual stimuli are shown on either side. The subject is instructed to shift attention to one side of the screen, without moving the eyes, when a visual cue, such as an arrow, indicates an upcoming stimulus at the side of the screen. Behavioral performance (speed or accuracy) in a visual discrimination task is enhanced when subjects shift their attention to the side of the screen containing the stimulus.

On the basis of studies of patients with attention deficits as a consequence of a neglect syndrome following a stroke, the parietal and frontal lobes have long been implicated in the control of visual attention (Chapter 17). Using PET imaging, Michael Posner and his colleagues have confirmed that frontal and parietal lobe regions contribute to the control of attention. Similarly, as we saw in Chapter 17, electrophysiological studies of attention by Michael Goldberg have identified neurons in areas of the parietal lobe that respond more strongly to attended stimuli than to unattended stimuli.

William James described two different kinds of attention. One is passive, automatic, stimulus-driven, and transient, whereas the other is active, voluntary, conceptually driven, and sustained. In "passive immediate sensorial attention the stimulus is a sense-impression, either very intense, voluminous, or sudden … big things, bright things, moving things … blood." We now refer to passive, nonvoluntary attention as exogenous attention, whereas active and voluntary attention is called endogenous attention.

Functional imaging has revealed that the two types of attention recruit different subregions of the brain. During voluntary attention certain parietal areas (within the intraparietal sulcus) and frontal areas (frontal eye fields) are active when subjects are instructed to shift or maintain attention. Shifts in attention are mediated by transient responses in the frontal and parietal regions immediately following the presentation of a cue. The maintenance of attention, critical for our ability to focus on a particular location in the visual field over an extended period of time, is mediated by sustained activity in the visual cortex as well as some of the same parietal and frontal brain regions. Additional brain areas become active when attention is diverted by a particularly significant or unexpected stimulus. For example, the amygdala is involved in diverting attention to emotionally salient (particularly fearful) stimuli such as a fearful face or a snake (see Chapter 48).

Attention is believed to be controlled by a particular network of cortical and subcortical areas. But how does this give rise to the improved behavioral performance discussed above? One idea is that signals from higher-order cortical areas flow back down to sensory cortical areas to facilitate the sensory representation of an attended stimulus. Functional imaging experiments have found that in this way the neural representation of an attended stimulus is amplified. This amplification is correlated with, and is believed to cause, the improvements in behavioral performance that accompany attention (Figure 20–7).

Despite its remarkable abilities, functional imaging, like any tool, has some technical and conceptual limitations. Four variables that correlate with brain metabolism can be used in functional imaging: glucose uptake, cerebral blood flow, cerebral blood volume, and deoxyhemoglobin content. The latter is the basis of the BOLD response on which most fMRI measurements are based (see Box 20–3).

Deoxyhemoglobin content is the only one that cannot be measured in absolute terms and is in fact dependent on a complex and poorly understood interplay among cerebral blood flow, cerebral blood volume, and the basal (or "resting") state of the brain region under investigation. Two brain regions with differing basal states, therefore, may lead to different BOLD responses, even if a stimulus induces identical changes in oxygen metabolism in each region.

Thus, inferring that differences in the BOLD response necessarily reflect underlying differences in oxygen metabolism and neural activity can lead to false conclusions. For example, the amplitude of the BOLD response to visual stimuli measured in the primary visual cortex is typically larger than the BOLD response to motor stimuli measured in the primary motor cortex. One might conclude that the visual cortex is metabolically more responsive then the motor cortex, but this difference between the regions is more likely to reflect differences in basal deoxyhemoglobin. The BOLD response, therefore, can be similar in several areas of the brain stimulated by a particular stimulus, but this similarity cannot tell us which area of the brain is more metabolically active than another.

Moreover, because most disorders of the brain affect the basal state and deoxyhemoglobin of targeted brain regions, similar false conclusions might be drawn when comparing the BOLD responses of patients and healthy controls (Figure 20–8). This limitation has hampered the usefulness of BOLD fMRI in clinical populations but can be overcome by calibrating BOLD measurements. Moreover, new functional MRI techniques measure cerebral blood flow or cerebral blood volume in absolute terms.

What component of neuronal activity is most highly correlated with these measures of brain metabolism? Although it is known that the BOLD response (and the other functional imaging measures) is triggered by metabolic demands of increased neural activity, the details of this process are poorly understood. Considerable evidence suggests that increased blood flow follows from increased synaptic activity. Thus, fMRI responses may be most closely related to synaptic input and intracortical processing within a cortical area, not the spiking output, and there are some clear demonstrations that change in blood flow can be dissociated from spiking activity.

Cortical circuits are, however, dominated by massive local connectivity; most synaptic inputs originate from nearby neurons, while only a small minority originates from distant sites such as the thalamus or other cortical areas. Thus, synaptic inputs in the cerebral cortex are produced mostly by neighboring neurons, leading typically to a tight coupling of synaptic and spiking activity, as well as metabolic responses (including BOLD fMRI). It is not surprising, therefore, that fMRI responses have often been found to be highly correlated with neural spiking. The extent of decoupling of synaptic and spiking activity depends on the nature of the cortical circuits, that is, whether the cortical activity is dominated by the local recurrent circuitry or by synaptic inputs (either feedforward or feedback) from other brain areas. Functional MRI measurements may be highly correlated with the spiking activity of a brain area under some circumstances, but with subthreshold modulatory input under other circumstances. One implication of this is that fMRI measurements must be interpreted with caution (Box 20–5).

Functional imaging provides a key bridge between behavioral studies of human cognition and electrophysiologic studies of neural function in intact experimental animals such as monkeys or genetically modified mice. Insofar as cognition emerges from a complex interplay across many regions of the brain, the ability to record brain activity simultaneously from multiple areas endows functional imaging with unique exploratory powers.

Functional imaging is expanding on three fronts. First, although it is clear that functional imaging can record meaningful signals from the brain at work, it is also clear that we do not yet have a complete understanding of what is causing these changes in signal, how they are generated, and what they are precisely telling us about underlying neural processes. Achieving a full understanding will require coupling functional imaging with invasive techniques. Accordingly, one of the important developments in the field is in the use of monkeys and transgenically engineered mice to investigate the physiological and molecular mechanisms underlying functional imaging.

Second, functional imaging studies are moving beyond establishing a catalog of the component parts of cognition toward an understanding of how the parts interact within large-scale networks. For example, investigators are beginning to explore how higher- and lower-order sensory areas interact with the medial temporal lobes and the parietal and prefrontal cortex to give rise to cognition or consciousness.

Third, studies are scaling down from the large-scale analysis of the whole brain to a more focused investigation of discrete brain areas, such as the numerous visual cortical areas, the olfactory bulb, and the hippocampal formation. These studies are designed not for human brain mapping but rather to test computational theories of the function and functional organization of predefined brain areas.