Brain Scans Don't Catch The Brain In Action

A visitor to the Wellcome Collection's 2012 exhibition "Brains: The mind as matter" looks at a functional magnetic resonance image (fMRI) showing a human brain as it listens to Stravinsky's "Rite of Spring" and Kant's third Critique.

 

The backlash has begun! After years of overselling neuroscience and its results in the popular media, we are now finally beginning to hear public voices of dissent. Alissa Quart, writing in The New York Times, warmly sings the song of the backlash, while Gary Marcus, over at newyorker.com, explains that the proliferation of "brain porn" — colorful brain scan images fatuously illustrating articles about you name it — shouldn't obscure the fact that understanding the brain is a worthwhile goal and one that is necessary if we our to understand ourselves.

It would be hard to overstate the extent to which the fervor about the brain-basis of human experience is stoked by the development in the last few years of new technologies for brain imaging.

Until very recently, post-mortem autopsy has been just about the only way to study a person's brain. The brain has remained, for science, a black box. At best we have been able to draw conclusions about its design and functionality by looking at what possessors of brains can say and do. Things are different now, or so it is widely believed. The development of PET, and more recently fMRI, enable us now finally to penetrate the black box.

Not so fast. A functional brain image such as those produced by PET and fMRI no more captures the brain in action than a graph illustrating the percentage of the population who go to church on Sunday captures the people in the act of worship. Brain scan images are graphical renderings of what we hypothesize is going on in the head. There's nothing wrong with such images. In fact, they are valuable tools for carrying on scientific study. But they are not pictures of our brains in action, and so they are positively not images of our minds at work.

To appreciate this, consider that we face a problem from the very beginning about how to decide what neural activity is relevant to a mental phenomenon that we want to understand. Scientists start from the assumption that to a mental task — say the judgment that two given words rhyme — there corresponds a neural process. But how do we decide which neural activity going on inside you when you make a rhyming judgment is the neural activity in which the mental act consists? To do that, we need to have an idea how things would have been in the brain if you hadn't performed the rhyming judgment; that is, we need a baseline against which to judge that the deviation from the baseline corresponds to the mental act. One way to do this is by comparing the image of the brain at rest with the image of the brain making a rhyming judgment. The rhyming judgment presumably depends on the neural activity in virtue of which these two images differ.

But the brain is never at rest! There are stages of sleep when your brain is working harder than it does at most times during the day.

The standard way forward is the method of comparison. For example, suppose you have a bunch of PET images of people listening to recordings of spoken words and then making judgments about whether given pairs of words rhyme. To isolate the area of activation responsible for the rhyming judgment, as distinct from the auditory perception of the spoken words, a standard procedure would be to compare these images with a second set of images of people listening to recordings of spoken words but not making rhyming judgments. Whatever areas are active in the first set of images, but not the second, would be plausible candidates for the place in the brain where the rhyming judgment happens.

But notice: the upshot of all this is not a picture of rhyming perception happening in the brain. It is an argument, and one that could turn to be mistaken.

To give an example of just one assumption at work in the reasoning: the comparison method assumes that there is no feedback between the neural activity that the brain is doing when we make a rhyming judgment and what the brain is doing when we perceive the words. If there were feedback, then it would follow that overlapping regions in the images do not necessarily correspond to a common neural factor that can be factored out. Now, as a matter of fact, it is highly like that there is feedback. There are neural pathways heading back into the brain from the eyes; but there are even more neural pathways heading back out again. And this should not be surprising. Consider how much easier it is to hear a sound that you are expecting than one that you are not expecting. This assumption that there is no feedback in the neural circuitry is the flip side of a different assumption that we can factor the cognitive act itself into (in this example) distinct, modular acts of perceiving the words (on the one hand), and judgments about whether they rhyme (on the other).

That's a substantive empirical claim about the character and composition of cognitive acts themselves and certainly not one that can be simply taken for granted. (I rely here on an excellent older discussion of assumptions in brain imaging: Guy C Van Orden and Kenneth R. Pap's "Functional neuroimages fail to find pieces of the mind in parts of the brain in Philosophy of Science 64, 1997: 85-94.)

I am using the rhyming case as an illustrative example. My aim is not to show that there is anything in the least misguided about the method of comparison. What I do want to bring out is that brain scanners don't simply show us what is going on when we listen and judge.

In a way, these considerations about feedback in the brain and cognitive models are only the tip of the iceberg. PET and fMRI have very low spatial and temporal resolution. When you localize events in the brain, using these techniques, you localize them to cubic regions of between 2 and 5 mm, that is, to regions in which there are hundreds of thousands of cells. If there is specialization or differentiation among these cells, that won't show up in the illustration. Nor, for that matter, can we be sure exactly when neural events are happening. Cellular events unfold at the scale of thousandths of a second, but it can take much longer time scales (large portions of a minute) to detect and process signals for making images. For these reasons, scientists have developed techniques of normalizing data.

Typically, data from different subjects is averaged. The averaging process involves the loss of considerable information. After all, brains differ from one another no less than faces do. Just as the average American tax payer has no height and weight, so averaged neural activity has no location in any particular brain. For this reason, scientists project their findings onto an idealized, stock brain. The pictures we see in Nature are not snapshots of a particular person's brain in action.

Finally, putting all this to one side, it is important to be clear that there is no sense in which PET or fMRI illustrations deliver direct information about consciousness or cognition. They do not even deliver direct representations of neural activity. Functional brain imaging systems such as PET and fMRI build images based on the detection of physical magnitudes (such as radio or light waves) that are are believed to be reliably correlated with metabolic activity.

For example, in PET, one injects a positron emitting isotrope into the blood stream; PET detects the emission of gamma rays caused by the collision of positrons and electrons. In this way, the PET image carries indirect information about metabolic activity (based on the direct measurement of a physical magnitude) which is in turn supposed to carry information about neural activity. The latter supposition is not unreasonable. Neural events require oxygen, and so they require blood. The neural activity, in its turn, is supposed to correlate to significant mental activity. Brain scans thus represent the mind at three steps of remove: they represent physical magnitudes correlated to blood flow; the blood flow in turn is correlated to neural activity; the neural activity in turn is supposed to correlate to mental activity.

If all the assumptions are accurate, a brain scan image may contain important information about neural activity related to a cognitive process. But we need to take care not to be misled by the visual, pictorial character of these images. Brain scans are not pictures of cognitive processes in the brain in action.