Neuroimaging in trauma studies is bad science and worse understanding

DSC00633In his recent book, The Body Keeps the Score (2014), Bessel Van der Kolk writes about the first time he used neuroimaging in his work. He placed eight traumatized subjects in a scanner similar to the fMRI (functional MRI), which records in real time changes in the blood flow in the subject’s brain. With the subject in the scanner, an associate read back an account of the subject’s traumatic experience(s). All subjects experienced flashbacks while in the scanner.

I taped the scans up on the refrigerator in my kitchen, and for the next few months I stared at them every evening. It occurred to me that this was how early astronomers must have felt when they peered through a telescope at a new constellation. (van der Kolk, 2014, p. 42)

fMRI: the basics

Is neural imaging really is equivalent to the Galilean revolution? In order to answer that question, it will be useful to understand how neuroimaging works in practice in the study of psychic trauma.

Unlike the MRI images most of us are familiar with from medical scans, which capture static slices of the brain or other organ, the fMRI measures contemporaneous blood flow in the brain, which is a proxy for neuronal activity. When a particular part of the brain, say the neurons in the amygdala, are excited, they require more oxygen and glucose, both carried in the blood. The data record an increase when an area of the brain is excited, reflecting the assumption that the neurological activity is drawing on oxygen in the blood, and a decrease when the excitation ceases, reflecting the assumption that the neurological activity has returned to its baseline level.

There is actually a complex statistical analysis behind the assumption (buried in most of the “fMRI analysis in a box” statistical programs) regarding the relationship between blood oxygenation and neural activity, including questions of time lag. For example, there may be a lag of as long as five to ten seconds between when neurons are excited and when they consume more oxygen, which is what is being measured. In addition, most neuroscientists assume a linear relationship between oxygen use and degree of neuronal stimulation, but that is a working assumption, not an empirically established fact. One reason to question this assumption is that the resolution of even the most sophisticated fMRI and its associated computer program is not high, about 27 mm3 (a cube with 3mm length sides). Today a few of the most powerful scanners can resolve down to the level of a 1mm cube, but they are not currently used in trauma studies.

This cube is called a voxel (a combination of volume and pixel), containing millions of neurons. The assumption that the blood flow being measured is not providing oxygen to more than one set of neuronal discharges is far from given. Furthermore, many millions of neurons have to be activated for a change in blood flow to be detected. Important brain functions may not require large amounts of blood flow to support them. Most studies assume this issue away as well. All studies make assumptions. The problem is that most published studies applying fMRI do not address these assumptions, leaving the average reader without a context to evaluate them.

Voxels light up when they are oxygenated, but they light up for various other reasons, some known some not, which we collectively call chance. A legendary study put a dead salmon in a fMRI and found quite a lot of neuronal activity, as measured by voxel light ups, over 130,000 per scan of the salmon “looking” at images of people (Craig Bennett won an Ig-Noble Prize for this study). Voxels, it should be remembered, are generated by the response of blood to an intense magnetic field. It should not be surprising that such fields generate a lot of noise, or disturbances in the field. The colors we see in pictures of a fMRI are probability estimates, rendered as colors rather than numbers, that a voxel (actually a contiguous group of voxels, organized into what is usually called a module, such as the amygdala, or other portion of the brain), was active by chance. Furthermore, it is the summation of the probability estimate of all subjects in the experiment, not just one. Colored images of the brain are essentially Excel spreadsheets rendered in the image of a single brain.

The quality of fMRI studies varies. Sometimes it sounds like phreneology, only this time done from the inside out. Temple Grandin (who writes about autism from the first-person perspective), suffers from panic attacks, explaining this fact in terms of how a MRI of her brain revealed that her amygdale, a brain region associated anxiety, was twenty-two percent larger than normal.

Other research has found large amygdale to be associated with having more friends and more complex social relationships, while still other research has associated amygdala activity with artistic creativity, as measured by responses to Rorschach test blot 03. In fact, talking about “the amygdala” is a little bit like talking about the brain, or rather a little brain in a big brain. The amygdala is a composite structure, composed of parts smaller than a 3mm voxel can discriminate. Whatever is going on with the amygdala, it is not likely to be a matter of sheer size.

It cannot be overemphasized that most of the action in the fMRI is in the statistical work that follows. Since the oxygen use of stimulated neurons is only slightly greater than baseline, which is itself not stable in or across individuals, sorting out the difference requires statistical legerdemain. Yet, studies using 12 subjects are routinely published; van der Kolk’s had 8. Finally, it should not be forgotten that oxygen use is only a proxy for neuroactivity, which is measured in microvolts. Several factors, including cerebral blood flow, cerebral blood volume, and cerebral metabolic rate of oxygen can affect the result in each individual and across individuals. The point is that there is a faint and sometimes nebulous signal, and a great deal of noise. There is no substitute for a large N (number of subjects). The minimum should be 40. Nor is there any substitute for a testable hypothesis, not one that is simply the restatement of the best data fit.

The meaning of neuroimaging

Even if the criteria of best science are met, the significance of neuroimaging, and CNS in general, is unclear. Consider the following example, originally published in The New Yorker. It is farcical but it makes a point. Patricia Churchland is addressing her husband Paul after a hard day at the office (the Churchland’s are both distinguished neuroscientists).

Paul, don’t speak to me, my serotonin levels have hit bottom, my brain is awash in glucosteroids, my blood vessels are full of adrenaline, and if it weren’t for my endogenous opiates I’d have driven the car into a tree on the way home. My dopamine levels need lifting. Pour me a Chardonnay, and I’ll be down in a minute. (MacFarquhar, 2007, 69)

If you think about it for a moment, you realize that this is really just an exercise in translation, everyday terms such as “well being” translated into serotonin levels. Translation isn’t explanation. Of course, no one is offering this as a real explanation, but it captures what much of applied CNS does, translating experiential mind terms into brain terms, not so different in principle from translating English into French.
More than this, the applied CNS translation is completely dependent upon, while studiously ignoring, the social and emotional context of the exchange, which is presumably something like this.

What she really wants is to express herself, and for her husband to care enough about her mental state to fix her a drink—not an East Coast martini but a varietal wine that almost defines California living—and give her some space—another stereotypically Californian request—to wind down. That is what the social interaction is all about . . . . The problem is that you cannot reduce the mental and the social to the neural without leaving something crucial out—namely, the mental and the social. And when you leave out the mental and the social, you have just kissed psychology (and the rest of the social sciences) good-bye. (Kihlstrom, 2010, p. 774)

And when you kiss the understanding of psychology and the social sciences goodbye, you’ve just kissed human understanding goodbye. Not because psychology and the social sciences are the greatest advances in human understanding, but because they understand themselves in human terms, using the language of human relations, even if it is often disguised by jargon.

We see that the really important issue isn’t whether one could explain Patricia’s feelings and desires in terms of her neurochemistry. Quite possibly one could. The claim is one of meaning. When we ask for an explanation of what Patricia is doing and what she wants, we are generally looking for an explanation at the same level of human meaning, for that is what an explanation means. This is not what we have to be asking for, but it appears to be what Patricia wants: a gesture of caring, a little space for oneself, some alcohol to help wind down, knowing that in a little while she will be sitting together with her life’s companion enjoying a good dinner. What could be more human than that?

Not the existence, but the widespread appeal and appropriation of applied CNS, particularly the fMRI image, as explanation of the human condition is a sign of the impoverishment of our culture. Not only is the neural image a reification of human emotion, but emotion loses its connection to human relationships to which it is bound.


Kihlstrom, J. (2010). Social neuroscience: The footprints of Phineas Gage. Social Cognition, 28, 757-783.

MacFarquhar, L. (2007, February 12). Two heads: A marriage devoted to the mind-body problem. New Yorker, 56-69


Leave a Reply

Your email address will not be published.