Can brain scans reveal how we think?
Scientists debate whether such higher-order processes as consciousness and morality have their own real estate in the brain.
Some people with alcoholism can change their behavior and remain abstinent, while others fight the battle over and over again. And while treatment for alcohol abuse is often effective, many patients wind up backsliding. If clinicians knew which ones are most likely to do so, they could intervene to help them stay sober.
In the not-too-distant future, a clinician in that position might turn to brain imaging for answers. Using functional magnetic resonance imaging (fMRI), a method that maps neural activity to specific locations in the brain, Yale scientist Dongju Seo, Ph.D., and Rajita Sinha, Ph.D. ’92, Foundations Fund Professor of Psychiatry, professor of neurobiology, and in the Child Study Center, examined 45 alcohol-dependent patients and compared the scans of those who later relapsed to the scans of those who did not. Surprisingly, when the relapse group tried to mentally relax, the prefrontal cortex failed to settle, and during stressful thoughts, it failed to activate, according to their study in JAMA Psychiatry in May. If your brain activity looks like that, you’re less likely to stay sober—or so, apparently, says the scan.
fMRI is the closest we can come to watching the brain at work. Its vividly colored images seem to offer snapshots of thought and emotion themselves. The central tool of many brain researchers at Yale and around the world, fMRI holds the promise of illuminating the brain-mind connection.
“It provides information that can’t be obtained with any other approach right now,” said Hal Blumenfeld, M.D., Ph.D., FW ’98, professor of neurology. The method is noninvasive and shows the whole brain at once, with better resolution in time and space than older methods can offer.
Certainly, fMRI’s ability to peek into our heads hasn’t been lost on lawyers, advertisers, and entrepreneurs. Brain scan findings have been used in court to defend sociopaths, while “neuromarketers” have used fMRI to measure audience reactions to a Harry Potter movie trailer. A company called No Lie MRI claims to have developed a reliable lie detector test—or “truth verification technology”—based on fMRI.
Yet as the technology comes of age, some observers of the field are calling for caution, and earlier this year two Yale authors published books arguing that fMRI is all-too-often misused. In Brainwashed: The Seductive Appeal of Mindless Neuroscience, co-author Sally Satel, M.D., HS ’88, a lecturer in the Department of Psychiatry, examines the implications that our hasty embrace of fMRI may have for the concepts of free will and human agency. Amid the popular enthusiasm for brain images, she argues, misunderstandings abound and dubious conclusions are often drawn. For example, when predicting an alcoholic patient’s behavior with fMRI findings, she said, we risk falsely concluding that relapse is inevitable.
In his recent book, Brain Imaging: What It Can (and Cannot) Tell Us About Consciousness, Robert G. Shulman, Ph.D., professor emeritus of molecular biophysics and biochemistry, questions whether fMRI should be used to study such higher-order cognitive processes as working memory, attention, and consciousness. A biophysicist who pioneered the technique in the early 1990s, Shulman believes that the design and interpretation of many studies that use it have been faulty. The brain, he argues, is best studied just like any other organ—via a physiologic approach that can identify neural processes that are necessary for a person’s behavior.
But many neuroscientists believe that fMRI can indeed get at higher-order functions—especially when combined with other measurement methods—and that research methodologies are improving, reducing the risk of unwarranted conclusions. They say that Shulman’s call to limit themselves to neurophysiology and behavior would do science and patients a disservice.
Present at the creation
Shulman was among the first physicists to study biological systems with nuclear magnetic resonance, and by the late 1970s, working at Bell Labs, he was using it to study how glucose is metabolized in yeast and muscle. That decade also saw the first magnetic resonance images and the first whole-body MRI scanner. Improvements in MR technology set the scene for the development of functional MR imaging at Yale and the University of Minnesota in 1992 (see sidebar: “BOLD beginnings”).
MR imaging had been a major advance in revealing anatomical structures. Functional MR went a big step further by mapping brain activity to specific locations and superimposing that data over the MR image. It exploits the propensity of hemoglobin to behave differently in a magnetic field—depending on whether or not it is oxygenated—a principle called blood-oxygenation-level-dependent (BOLD) imaging. Because active neurons consume oxygen, the brain compensates by sending oxygen-rich blood their way; fMRI can map areas of neuronal function by tracking the flow of oxygenated hemoglobin.
After the initial studies in 1992, scientists rushed to adopt fMRI, finding it a faster, more accurate, and more accessible way to image brain activity than such older technologies as positron emission tomography, or PET. Early experiments yielded detailed, reproducible maps of brain areas corresponding to sensory and visual stimulation. The technique has been central to the advent of cognitive neuroscience, a developing field that studies the neural basis of higher brain functions. Cognitive neuroscience studies have implicated the amygdala, for example, in evaluating threats and mediating emotional learning. Circuits in the hippocampus appear to be critically important for relational memory, which allows us to associate names with faces. Parts of the prefrontal cortex seem to power down in schizophrenia, and so on. Previous methods had produced a great deal of information about the functions of these and other brain areas, but fMRI allowed scientists to ask more sophisticated questions and clarify what they previously had only suspected was true.
In 2000, for example, investigators at Yale’s Child Study Center found evidence, through fMRI, that subjects with autism don’t process faces in the brain’s facial recognition center. Instead, they use an area of the brain normally associated with recognizing objects. The finding caused a sensation in the autism community—it seemed to explain why autistic children tend to show little interest in faces. Their 1997 work followed studies by a group at Yale led by Gregory McCarthy, Ph.D., and by another group at Harvard and Massachusetts General Hospital led by Nancy Kanwisher, Ph.D. She and colleagues including Marvin M. Chun, Ph.D., now a professor of psychology at Yale, showed that this brain region—the fusiform face area—is selectively activated by faces, confirming years of suggestive but ambiguous data from other methods. It is today the most-cited fMRI brain research paper in the scientific literature.
Where morality and memory reside?
Shulman believes that many fMRI studies are too ambitious. Mapping brain areas specialized for sensory or motor systems is one thing. Mapping the life of the mind is quite another. Memory and attention are subjective processes that cannot be experienced by an observer, and they may not be as discrete as we think they are. We talk about remembering to pick up the kids or remembering a phone number, but those two acts may not be as fundamentally similar as our single term for them would imply. “When you start looking for localization of concepts like honor, values, morality, memory, consciousness, you aren’t going to find them,” Shulman said, “because we have never learned exactly what they are.” Shulman points to a UCLA study purporting to show that Republicans have higher amygdala activation and are more likely to vote based on fear and other emotions. Such experiments, he said, constitute “phrenological fMRI,” a term critics have used since the early 2000s to dismiss such research.
To grapple with such objections, it’s important to understand a few things. For one, the brightly colored images that appear in journals and news reports usually don’t represent one brain at one time; rather, they represent highly processed, composite results obtained by processing several individuals’ brain data through statistical algorithms (see sidebar: “How functional MRI works”). Moreover, these algorithms rely on assumptions not everybody agrees on.
Second, BOLD imaging has important limitations. Though increased oxygen-rich blood and its stronger BOLD signal usually flag increased neuronal activity, there’s a time lag, since neurons fire thousands of times faster than blood flows. Moreover, sometimes the BOLD signal is positively misleading. Yale’s Blumenfeld and his colleagues were the first to show that, in some seizures, neurons fire in such a frenzy that they need more oxygen than the brain can deliver and the BOLD signal goes down instead of up. They also found that blood flow sometimes declines in response to neuronal activity. These findings strike at the heart of all BOLD assumptions. “Taking BOLD alone is always going to be potentially risky,” said Blumenfeld. “Everyone is hoping for [better] techniques. ... I’ve been hoping for it my whole career.”
Third, and perhaps most importantly, experimental premises are crucial. Cognitive neuroscience assumes that mental processes like working memory, attention, problem-solving, and decision-making are real, objective, measurable, observable phenomena. Especially in the early years of the field, cognitive neuroscientists believed that these brain functions reside in discrete modules, a school of thought called localizationism. Many researchers have come to believe that these functions are organized in networks. Others posit that the whole brain is involved in all functions—the aggregate field view (see sidebar: “ ‘Mixed-use development’ in the brain”).
These things matter because they affect how a researcher plans and interprets experiments. A localizationist who expects working memory to reside in one particular spot in the prefrontal cortex will naturally process his data to look for that area lighting up. But there are other ways to analyze the same data set that can lead to different conclusions about which areas of the brain are active during a cognitive task.
Chun said he believes that Shulman “has appropriately urged caution over the years, but his concerns do not acknowledge all the recent advances in analysis methods that enable more precise interpretation of BOLD signal activity for understanding perception and cognition.” As an example, he points to the work of Jack L. Gallant, Ph.D. ’86, at UC Berkeley. Gallant’s group has produced highly complex, interactive brain maps, derived from enormous datasets that attempt to correlate the neocortical activity of study subjects with hundreds, even thousands of objects and actions observed by the subjects. The resulting images and word maps, when viewed dynamically on a computer, are far more nuanced than the 2-D brain slices that have become familiar since the first fMRI studies in the early 1990s.
A change of mind
Shulman recalled that shortly after the development of functional imaging, the idea of modules for memory, consciousness, and other cognitive concepts raised hopes that finding where they reside would explain them at last. “Well,” he said, “that did not work.” Initially excited by the promise of fMRI to explore cognition, by the mid-1990s he had conducted an experiment that changed his mind. He showed that a certain region of the frontal cortex lit up during a task of working memory. After publishing his results, he realized that he hadn’t demonstrated that this response was unique. Repeating the experiment with an attention task, he found that the very same area lit up—a contradiction of the assumption that different mental activities occupy distinct, nonoverlapping modules in the brain. At the same time, metabolic studies showed that even at rest the neuronal activity of the brain is very high. An absence of change in activity during a task did not mean a brain region was not involved in supporting it; instead its activity could just be the same in the task and control states.
Shulman had committed the reverse inference error—working backwards to link activity in a brain region to a specific cognitive function. This error is one that he and cognitive neuroscientists agree has led many fMRI researchers to overstate their results. (In contrast, Chun’s 1997 paper on the fusiform face area asserted that it is selectively activated by faces, a conclusion drawn after comparison with various control stimuli.) Shulman came to believe that the very philosophical underpinnings of such experiments are shaky, since they assume a modularity that isn’t neatly borne out by the findings. Context is all-important: how working-memory tasks look under fMRI varies widely, depending on the nature of the task.
A more effective use of fMRI, argues Shulman, would be to characterize the brain’s activity during observable behaviors in brain imaging studies. “For example, the total brain activity necessary for a person to perform the act of memory can be observed,” he said. “The location of the psychological concept of memory cannot.”
Critics’ concerns aren’t limited to experimental method; like Satel, some argue that results are being exaggerated. In an opinion piece in The New York Times the authors of the UCLA politics-related study claimed that fMRI results revealed how 20 voters felt about Hillary Clinton and Mitt Romney. Exasperated neuroscientists at a dozen universities responded in a letter to the Times—it’s not possible, they said, to determine a person’s mental state by looking at a brain scan.
Contrary to hopes, Satel argues, the technology cannot sway voters, sell products, sniff out lies, or reveal the causes of crime and mental illness. “To regard research results as settled wisdom,” she writes, “is folly.”
Satel views neuroscience and its tools as nothing short of remarkable. But she thinks that we are too quick to believe that this young science has at last illuminated the mind-brain relationship. Our rush to explain complex behaviors via brain activity alone fails to take psychological or social factors into account—and can lead us astray. She is skeptical, for example, of the way that fMRI findings have been used to argue that addiction is a purely a brain disease. Moreover, she writes, “the fact that addiction is associated with neurobiological changes is not, in itself, proof that the addict is unable to choose.” Recovery programs that make use of incentives and consequences work for addicts, she pointed out, but would never help a Parkinson patient.
The brain’s integrated networks
Like psychologist Chun, Marc N. Potenza, Ph.D. ’93, M.D. ’94, believes that our understanding of brain organization is outgrowing its initial simplicity. Potenza is a professor of psychiatry, neurobiology, and child study who uses fMRI to study behavioral addictions like compulsive gambling.
Conventional wisdom, he said, held until recently that the amygdala processed fear, the ventral striatum provided drug-induced rewards, and so on. But these brain regions have been implicated in other processes as well, pushing cognitive neuroscience toward a network-based model. “The way in which these regions work together in networks or functionally integrated activations that some MRI data can identify, that’s really important,” Potenza said. The Human Connectome Project, in which research universities share fMRI data on brain networks, is a first attempt at mapping such connections.
Adoption of a network model isn’t the only shift in thinking. Researchers are using fMRI results to break traditional concepts like working memory into such smaller and more isolable components as encoding, shifts of attention, and retrieval. They are also studying the default mode network, brain activity when a person is awake but not doing anything in particular, using both cognitive and physiological approaches.
There is also brain plasticity to consider. Existing functional connections in the brain can be readily altered through learning and experience—and we can see those changes on fMRI.
Potenza rejects the idea that only behaviors observable by others constitute the proper subject of fMRI study. “There are some conditions like major depression where subjective accounts are very important to understand,” he said. “If we were to omit looking at subjective responses, motivational states, emotional states, we would be limiting ourselves with respect to our understanding of the human condition in multiple clinically relevant states.”
Perhaps some of today’s popular fMRI applications will recede into history, taking their place alongside early 20th-century electrical healing gadgets and shoe store X-ray machines. Satel believes this burst of exuberance, if sometimes troubling, is normal in these early days of contemporary brain science. “You start out a little more crude, and then you perfect and perfect and perfect,” she said. “Wherever we are in 20 years, I doubt we’d be there had we not gone through this phase first.”