Brain Babies and Bathwater

In Martha J. Farah’s critique on functional magnetic resonance imaging (fMRI), she encourages fMRI skeptics not to “throw the baby out with the bathwater.” While she admits that fMRI does not read minds, she argues that it is still a viable method to inform cognitive theories, locate modular-connections that instantiate higher levels of cognition, and qualify localization with further explanation. However, one must first understand the limitations of both the technology and our ability to analyze the data retrieved. Farah promises a healthy level of skepticism to critique the criticisms, and supplies her “Kernel of Truth” within four different categories of the skeptics’ concerns. 

    In continuing with her fascination with alliteration, Farah first identifies the “gap between the neural events being studied and the images that purportedly represent them.” This she titles “Blood Vs. Brain.”  With a history of pseudoscience that leaves many scholars jaded, fMRI imaging is often dismissed as “hoopla” and criticized for its dependence on hemodynamic changes which some believe are themselves fundamentally flawed. She acknowledges these criticisms fMRI has cultivated in the field of cognitive neuroscience, and evaluates them for their worth. “fMRI is not a direct image of the brain,” says Judith Horstman. Blood oxygenation levels are indirectly correlated to neuronal activity through a series of high definition magnetic imaging; it is a fallacy, some argue, that affect (increased neuronal activity specific to the region of blood flow) can be assumed through these measures. Farah argues that little of what we call science involves direct “observations of the subject matter of interest.” This phenomenon is nothing new. Furthermore, “Complaints that functional neuroimages do not “show” brain activity appear to be based on a naive view of science and its methods.” Supporters of fMRI argue that correlation is not equivalent to meaningless. Therefore, just because one does not know the intermediate steps between cause and effect, it does not mean that there are none. 

    Late pop singer Michael Jackson remarks that “It don’t matter if it’s black or white.” fMRI skeptics, in terms of scan representations, boldly disagree. “Fabrication,” is a job losing word in the scientific research community; and often, the brain images produced by a research team look far more like the childhood game of “color in between the lines” than a realistic scan.  Farah and other critics argue that “the [fMRI] images are more researcher inventions than researcher observations.” Scientists and nonscientists alike have regarded “the safe use of color coding [certain modules in the brain] with suspicion.”  BOLD is not just an acronym derived from the data, but what they do visually to the images that are then available to the public eye. Images often appear like country maps, bordered to perfection, when in actuality the lines do not exist. There is not one area that performs a single cognitive function. Realistically, “differences in activity levels” (that the brain scans depict) “are tiny.” Some argue that scientists prejudicially use a color scale that favors warmer, brighter colors for representing higher activation. She relates this to the common scheme of “plotting numerical data on axes where higher numbers appear higher on the page” or adjusting a graphical scale to small changes in the Y axis look huge. Does Farah think so? Not at all. She believes that the bold colors depicting relatively tiny changes in the dependent variable are just for the convenience of the reader, much like if the Y axis of a temperature graph spanning only two degrees Celsius makes the “relevant relationships among data points salient.” Again, Farah comments that these issues are not unique to brain imaging, and so therefore neither should the critiques. 

    Phrenology is pseudoscience prevalent before humans were capable of studying living brains and is comparable to roman blood letting in efficacy. In a similar way, “functional neuroimaging has been criticized for encouraging research aimed merely at localizing psychological functions, for being incapable of testing psychological theories, for assuming a modular relation between mental and neural systems, and even being “caricatured as a form of phrenology.” Farah suggests taking a closer look into the benefits of how understanding localization in already well known areas of the brain (“on the basis of lesion studies or single cell recording in animals”) can inform us about areas still in the dark. 

    “Localization is merely a foundation,” says Farah. In building a house, one would not merely stop at identifying which tools are needed. Similarly, the functionality of fMRI does not stop at identifying which brain areas represent corresponding cognitive tasks.  Farah supplies a confident slew of evidence toward rejecting the “neo phrenology” charges coming from skeptics. She begins with a controversial debate between in the visual system: viewpoint-dependent representation vs. viewpoint-invariant representation. By using fMRI scanning methods and taking advantage of the “adaptation paradigm,” researchers can further distinguish our retina’s methods to decipher and recognize objects. Functional connectivity is the connectivity between brain regions that share functional properties. fMRI can utilize localization to pinpoint modules working coactively. These functional network of areas are complex and may change depending on the task conditions. Multivoxel pattern analysis allows researchers to gather a more detailed map of the intermediate processes from stimulation to encoding to information. In no way does this fit the description of “neo phrenology,” Farah argues rather triumphantly. 

    In further support of her belief that localization is merely a starting point, Farah begins to digest the area of localization in relevance to psychological theory. Similarly to how reaction time is not solely tested to analyze cognitive speed, fMRI is not constrained to the study of localization. The most difficult experiments to design are those involving psychological hypotheses. Farah acknowledges fMRI flaws in this area, however adopts Max Coltheart’s dismissal of these critiques. “A fairer and more realistic question is this: Can functional brain imaging contribute to confirming psychological hypotheses in roughly the way behavioral studies do?” This evens the playing field. A study identifying whether “the visual system does ‘double duty’ for perceptual processes and mental images generated from memory” confirms that fMRI can in fact confirm psychological hypotheses or at least provide progress. 

    Apart from localization, Farah begins to dissect another formidable claim. Neuroimaging shapes the way that we make hypothesis by introducing two problems, writes Uttal. Firstly, it ‘invites us to focus on a subset of the relevant data. Secondly, the mistaken idea that when all lesser peaks are reduced to invisibility by arbitrary scaling, the largest remaining peak represents the sole locale ofa particular cognitive process.”  In actuality, there is no “sole locale” of a cognitive processes. Hypotheses should be made on the microlevel, not the macro. The kernel of truth in this criticism is that early research focused on small macroscopic areas without considering iteration and functional connectivity. Hypotheses, Farah remarks, are truly selected in part based on research with other methods of neuroscience and psychology.”

    Inferring the correlation of one psychological process with subsequent brain activity is not a straightforward process. Since the brain is not a purely modular network, more than one cognitive action can produce similar activity in similar regions of the brain. Wanton reverse inference is a fallacious method of research pseudoscientists often use in which they begin with a scan and say that certain activity was caused by one specific cognitive action when in actuality, a large pool of actions could have the same effect. Some examples of this are brain -based lie detection and psychiatric clinics basing diagnoses on the premise that certain patterns of activation can be used to infer the presence of certain disorders.” Reverse inference is much more effective when paired with forward inference. Forward inference is a more logical type of assessment where a psychological or physical phenomenon is specified and then performed under scanning to see which part of the brain is activated. This is a one to one conclusion. Contrarily, reverse inference can only be made “with confidence when one knows the full range of psychological processes that could produce a given pattern of activation under the circumstances of the study.” This is highly improbable. Yet, once again these critiques are not specific to the field of neuroscience nor the technology of fMRI. 

    More skepticism comes from the large amount of statistics computed to average the data into an appropriate measurement. Statistics, like most other things, can be done correctly or incorrectly. Carole Wade argues that using statistics to blur individual differences in brain anatomy which “may make the ‘uniqueness of fingerprints or facial features seem simple by comparison,” is problematic when many subjects’ scans are aggregated to produce a single image. According to the nature of statistics, extensive use provides equally extensive probability for error. One of these potential errors arises through multiple comparisons, a problem specific to neuroimaging (that’s a first.) Imaging techniques utilize three dimensional voxels that are generally 3mm cubed. Each of these voxels “could be the site of an independent statistical test comparing the value of the BOLD response measured in that small bit of the brain between the conditions of the experiment.” If the threshold for activity of these voxels are too small, then the whole brain will light up and then researchers get to reverse infer whatever cognitive action they desire to whichever area the brain pleases them. That is why an acceptable P level of P > .001 has been established throughout the scientific community. One method to defray the error of this method is to set a priori region of interest - to “simply limit the number of comparisons by specifying in advance the regions relevant to the research hypothesis.” In a similar manner, researchers can identify which voxels are most activated by the stimulation, and then only test comparisons for those voxels. This leaves them with a much higher rate correlation effect than otherwise. The result is that the second round of data has been “enriched” by chance effects from decreasing the scope of the experiment and using points that are only inline with the hypothesis. After analyzing these critiques Farah again shouts her main disclaimer that this problem “is not unique to functional neuroimaging.”     

    The criticism that brain images are overly appealing or “unduly persuasive” are generally targeted at the scientific illiteracy of the public rather than the images themselves. Ironically, the are tailored so that the general public can easily receive them. There is harm, however, in the potential for “brain scan images to create biases in the laboratory, the clinic and the courtroom.” Also there is danger in the fact that viewers respond less dubiously when “scientific” claims are paired with brain images. Through several examples of jurors being presented information about the psychiatric state of defendants along with brain scans, Farah explains that this bias is actually unclear. She argues that “neither study found an effect of brain images over and above information delivered verbally.” 

    Neuroscience may possibly receive more credit and interest than it is scientifically worth, argues Farah. It is important to main goal-orientated and to use criticism as a mechanism toward progress. Because “a substantial minority of research is [stained with these ill-mannered methods],” these criticisms are both valid and useful. Farah argues that none of the criticisms discussed in her article constitute reasons to reject neuroimaging, but remind us that “like all other scientific methods,” it’s not perfect.  

    Another thing that is not quite perfect is Farah’s review. At the root of fMRI is the hemodynamic changes measured through magnetized oxygen molecules in the blood stream. Therefore, the most important components of testing are spatial and temporal resolution. Farah admits that “fMRI informs us about activity only in a relatively large area of brain tissue . . and can inform us only at relatively long time intervals. If we are not receiving data from enough time intervals to make the claim of cognitive causation, then fMRI cannot fully be trusted. Farah deflects strong critiques about signal noise and does not suggest the implications of multivariate pattern analysis.