STATS ARTICLES 2012
Neuroscience Or Neurobabble?
Rebecca Goldin, Ph.D., Cindy S. Merrick, July 16, 2012
In the first part of an ongoing series, we look at functional magnetic resonance imaging, and whether it’s really the window on the mind that some in the media – and science – would have us believe.
With the news media telling us that neuroscience – and brain scans – can explain everything from a global pandemic of Justin Bieber fever to whether you are likely to stay with your spouse, we investigate what neuroscience can and can’t tell us about who we are and why we do the things we do. In the first part of an ongoing series, we look at functional magnetic resonance imaging, and whether it’s really the window on the mind that some in the media – and science – would have us believe.
Gone are the days when the only people who believed in technologies that could read minds were distinguishable from the rest of us by their tin foil hats. With the advent of functional magnetic resonance imaging (fMRI), we are able to see, in near-video quality, the ebb and flow of a live mind at work. Or so it seems. Something, for certain, is at work, and there are lots of people willing to tell you they know exactly how to interpret what we can see. Certainly this new technology has already produced fascinating results: surgeons use it real-time to avoid critical regions while operating on brain tumors; physicians use it to look for changes in the brain activity of stroke victims as they experience physical rehabilitation; and fMRI data showing activity in the brains of patients thought to be in a vegetative state may be blurring the line that defines consciousness. Along with these advances, though, have appeared many somewhat less credible stories. The media reports claims ranging from fMRI’s ability to detect lies to its predicting future addictive behavior or determining whether or not you really love your spouse, or, maybe, your iPhone. Already, attempts have been made to use fMRI as admissible evidence of lie detection in court (so far, they have failed); and in another court case, fMRI results and a neuroscientist’s testimony were admitted in the sentencing hearing. The data were used as evidence that the defendant, a violent offender, was psychopathic.
Like a lot of promising new technologies, fMRI is in its infancy; but this has not stopped a great deal of fiction preceding the science, in large part due to the media’s fascination with the idea that we can now see how the mind works. But, unfortunately, media hype isn’t the only thing responsible. The rush to produce newer, better results in behavioral sciences research has brought a glut of studies lacking rigorous methods and analysis.
What is fMRI?
Functional Magnetic Resonance Imaging was developed as recently as the 1990s, when Seiji Ogawa and a team at Bell Laboratories experimented with mouse brains. The basic idea is that there are small differences in the local magnetic fields among different regions of the brain, depending on the activity (and, specifically, the resulting oxygen level) of those regions.
The method to measure these differences is a stroke of genius. It relies on the basic principal that hydrogen nuclei behave differently when there is a lot of oxygen around, compared to when there is not. Blood transports oxygen according to the need of specific regions in the brain. Hydrogen can be found abundantly in water molecules throughout the brain. FMRI measures differences in blood oxygenation levels in different regions of the brain by observing the reaction of hydrogen nuclei under certain physical conditions described below. Conclusions about brain activity itself are interpretations of these differing oxygen levels.
The mechanism for fMRI is as follows. A large magnetic field is exerted on the brain, resulting in an alignment among the protons in the hydrogen nuclei. After the large magnet aligns the protons, a radio frequency (RF) signal is sent from the machine to knock the hydrogen protons out of alignment (they are then said to be “phased”). When the RF signal is terminated, the protons then return to their aligned state (known as “dephasing”), emitting energy in the form of their own RF signals as they settle. This process occurs over time, and when the magnetic field is inhomogeneous, that is, containing irregularities, the signal is weaker, as nearby hydrogen atoms emit signals that cancel each other out. In contrast, a more homogenous, or uniform, magnetic field results in a stronger signal. The presence of hemoglobin without attached oxygen (also called deoxyhemoglobin) results in a more inhomogeneous magnetic field, and therefore a weaker signal.
The important point here is that fMRI does not directly measure neural activity; it measures oxygen levels that occur a few seconds after nearby neurons fire. All neural activity is inferred based on a statistical analysis of a series of measurements over time of the blood oxygen levels. In a typical fMRI scientific experiment, a person lies in the scanner while observing visual images, performing a small movement like finger-tapping, or receiving other kinds of stimuli. The fMRI machine meanwhile captures the RF signals sent from the brain. Data analysis involves looking for regions of brain activity that correlate with the stimulus. In some studies, the resulting statistical analyses of the brain scans are compared against scans of the same people not doing these tasks. For other studies, the scans of different people are compared – for example, one might compare the brain scans of people with autism against those without autism, checking for observable anatomical differences, or compare the scans of people making food choices while tired or not tired to see which brain regions may be involved with those choices.
Correlation vs. Causation
While the technology gives us a way to see what is happening — specifically, oxygen is being delivered in response to neural firing — it cannot tell us why. How should sudden flurries of activity be interpreted?
For example, fMRI scans of cocaine addicts found that their anterior cingulate cortex, a region known for emotional processing, was activated if they watched videotapes with “cocaine-associated cues.” This was not true of non-cocaine addicted subjects. It could be that cocaine use impacted these addicts’ brain activity, causing a different response to the stimuli than that of healthy people. It could also be that these brain differences were in place before the addicts began using cocaine, and the activity of the anterior cingulate made them vulnerable to addiction. A single fMRI study will not reveal the direction of causality (if indeed there is one) but simply notes the correlative differences in activity.
Mistaking correlation for causation is driving many false interpretations of fMRI data, both in media coverage of the topic and in the scientific community. Analysis of fMRI data can be used comparatively to observe differences in the brain activity elicited by different stimuli, but it does not provide a method to interpret the cause(s) of the activity itself.
Using fMRI data to identify individual needs/situations may be extremely successful even if causality cannot be assumed. For example, fMRI scans can indicate a person’s likelihood to quit smoking successfully. In one experiment, heavy smokers who desired to quit were observed with an fMRI while they watched advertisements designed to help them quit. Those who had more activity in their medial prefrontal cortex were subsequently more successful quitting. While the correlation was strong enough to predict which individuals would subsequently quit, there is no evidence that inducing those brain patterns is would provide a person with the tools to successfully quit. In other words, the brain patterns themselves may not be a causal factor in the ability to quit.
In an opinion piece for the Bulletin of the European Health Psychology Society in December 2011, Tal Yarkoni, a postdoctoral fellow at the University of Colorado at Boulder, notes that we don’t really know what such a correlation means on its own. Brain activity in quitters may be different from the others for a variety of reasons. The differences in activity could be pointing to fundamental differences existing among individuals before they even took the test. A successful quitter may be someone with greater motivation to quit at the time of the experiment, or is more thoughtful about how to quit. Such emotions or personal resolve may lead to differences in brain activity as well as differences in smoking cessation rates. While there is a clear correlation between more prefrontal cortex activity and increased success at quitting, it may be that the cortical activity is not the reason for success; it’s just a predictive marker.
The confusion between correlation and causation can, however, play into fears about common activities such as video gaming. In 2006, Rene Weber, et al., published a paper with the provocative title “Does Playing Violent Video Games Induce Aggression? Empirical Evidence of a Functional Magnetic Resonance Imaging Study.”
In their study, the subjects played a violent video game, while the fMRI machine recorded their brain activity. The authors noted that the brain activity patterns of the subjects during times of virtual violent action were the same as the activity “considered characteristic for aggressive cognition and behavior.” From this they conclude that it is the violent gaming experience that induces aggressiveness during play.
Yet a similarity of brain patterns in those who play violent video games to those who express violent or aggressive desires/intensions is hardly indicative of whether people who play these games will also tend toward violence in real life. There are many questions that arise about how to interpret this kind of brain activity. Specifically, would anyone (gamer or non-gamer) experience a similar brain pattern when playing these games? The study didn’t do a control, so we don’t know. But if so, one could argue that the gaming may induce these neural patterns (suggesting that the games recreate an aggressive situation, or ask the player to engage in aggressive behavior, which seems relatively “obvious”). If not, one could argue that people who have aggressive tendencies also enjoy violent video games, which, again, is not particularly surprising.
This brings us to the question of the test subjects themselves. The subjects were recruited using advertisements in video gaming and computer stores. The group tested consisted of 13 males between the ages of 18 and 26 who did, on average, 15 hours of gaming a week. In other words, this group is hardly representative of all people. Like the smokers in our previous example, the subjects may have been preconditioned for the brain activity the researchers observed – in other words, the observed brain activity may reflect their disposition to play long hours of video games in the first place. The initial (pre-gaming) brain state is not controlled for in the study, and thus calls into question their causal interpretation of the data (that gaming leads to aggression). And without evidence that these kids actually engaged in violent or aggressive activity outside their virtual world, we have no evidence that the video games are even correlated with youth violence. With ninety-seven percent of U.S. teens playing video games, violent kids are almost surely also playing video games; it hardly implies the gaming is at issue. Violence in video games may indeed lead to serious consequences — new evidence regarding the controversy is emerging all the time — but a study noting similar brain patterns does not make the cut.
Multiple Causes for the Same Result
In 2003, a woman was jailed for murder after the prosecution successfully argued that she deliberately poisoned her own child. The woman persisted in her denial, and throughout her jail term she and her family sought to overturn her conviction. What if technology existed that could prove her innocence by showing that her version of events is the truth, and the prosecutor’s is false? Certainly such a tool could revolutionize the criminal justice system.
In an interesting case study, published in European Psychiatry in 2007, Sean A. Spence, et al., studied fMRI data collected from a woman four years after her murder conviction in the death of her daughter. The experiment sought to determine if fMRI could be used to predict when the woman was answering questions about her guilt or innocence truthfully. Evidence of truth or deception would be based on two things from existing research using fMRI to study deception: certain areas of the brain are more strongly activated during deception than during truth-telling; and deception incurs greater overall brain activity. The authors, after studying details of the woman’s case, composed a list of 36 yes or no questions on the subject. On various scans, the patient was alternately asked to answer the list of questions either truthfully, or falsely, according to her belief, while the scanner recorded her brain’s responses.
The data gave a fairly consistent result: the activity believed to indicate truth telling correlated with her being asked to indicate her innocence. When she was asked to give responses that indicated guilt, they saw the expected deception patterns.
Can we conclude that the woman didn’t poison her daughter? Spence, et al., are very careful in describing their results. While stating the importance of findings that seem consistent with our current understanding of deception and brain activity, a full 25 percent of the paper is given to caveats that could plausibly lead to a “wrong” conclusion about the woman’s guilt or innocence. There are many questions for this experiment, casting doubt on its ability to determine whether she is telling the truth. For example, as the authors ask, “is it possible that the subject might have become so used to the details of her case that her responses are ‘automated’, so that they do not require the engagement of the cognitive executive to execute deception?” Keep in mind that this experiment was done four years after the woman’s conviction. Having spent that time fighting her conviction could clearly involve constant rehearsal and reiteration of an alternate (false) version of events.
Even more compelling is the question of whether some brain activity is actually an emotional response to the nature of the material. In a 2009 paper entitled “Great Expectations: What can fMRI research tell us about psychological phenomena?” Tatjana Aue and colleagues used Spence’s experiment to exemplify the need for caution in fMRI interpretation. To draw the suggested conclusion would be doing what they describe as failing to “consider the possibility that other antecedent conditions produce the same regional brain activation.” As Aue argues, “admission of a culturally reprehensible behavior such as harming one’s child may trigger strong control processes in an attempt to avoid detection.” In his own previous research, Spence has found that greater activation in key areas is caused when telling the truth about material that is shameful or embarrassing.
Finally, in the end, there is no accounting for the possibility that the woman simply believes she is innocent, even when she is not. The woman has been diagnosed with a disorder called “Munchausen Syndrome By Proxy,” which is characterized by a person who harms one in his or her care, usually by inducing illness, out of a desire for attention. Self-deception could be involved in the mental state of a woman who poisons, and ultimately kills, her own child. What’s more, very little is known about the neurobiology of denial and self-delusion, except that people do sometimes experience brain patterns demonstrating self-delusion.
Aue et al. suggest various ways to construct falsifiable hypotheses in order to reduce such errors of inference. Testing two or more theories that give differing explanations for the same outcome is one way to use fMRI data more definitively to draw valid conclusions, or at least rule some others out. They describe an example in which brain activity during object recognition is tested. The question for researchers was whether the difference between a thing seeming familiar and something definitely remembered was simply one of memory strength, in which there was a single mental process during both types of recognition, but the familiarity signal was weaker than the remember signal. A contrasting suggestion was that in fact two separate processes were responsible for the two different kinds of recognition. Such a hypothesis would be affirmed by observation of activity in separate areas of the brain during the two different memory types. The two mutually exclusive hypotheses can be tested against each other. Results from fMRI in this case are capable of clarifying which hypothesis was more likely the correct one, through affirmation of one and falsification of the other.
The most important point made by the Munchausen case is that, without the certainty that the specified brain activities related to lying and truth-telling are uniquely caused by those behaviors, we can’t say that fMRI alone predicts deception. It may be the case that, using this technology, deception patterns are indistinguishable from those of shame, embarrassment, or other forms of psychological distress. More comparative studies with deception patterns, such as the kind described by Aue, could better illuminate this area in question. In general, continued observation of an expected brain activity isn’t sufficient evidence that there do not exist other plausible cognitive states that appear to stimulate the same activity, which would lead to completely different interpretations of fMRI data.
FMRI technology continues to mature, with new techniques in data collection and sorting advancing rapidly. Ever more detailed close-ups are made, subdividing the brain again and again based on more finely-grained observations. Perhaps even more importantly, more critical views continue to be taken by neuroscientists at the exaggerated or over-optimistic claims from within their own community. No doubt the full potential of fMRI technology still lies before us, as does the promise of unlocking the secrets of the human mind. But fundamental flaws in reasoning still cloud many results, as the thrill of discovery upstages the necessary rigor and skepticism due every scientific pursuit.
Rebecca Goldin is the Director of Research of STATS at George Mason University.