The Dirty Secrets of fMRI

The blogosphere has begun debating the merits of fMRI. That's a good thing. The debate began with Paul Bloom's excellent editorial in Seed, in which he argued that "fMRI imagery has attained an undue influence, and we shouldn't be seduced." It continues here and here.

I used to work in a neuroscience lab grounded in molecular biology, and there was no shortage of fMRI bashing. A typical complaint went like this: "I'm here struggling with my damn Western blot [or Southern, or PCR, etc.], trying to be a good reductionist, while all those fMRI researchers just pick a sexy question, stuff some people in a magnetic tube, and get their technicolor pictures on the cover of Nature. It isn't fair."

Of course, that complaint wasn't fair either. fMRI isn't that easy. But I also wonder if there isn't a germ of truth behind the kvetching. Molecular biology is hard because reductionism is hard. It isn't easy to parse life into its parts, to break a brain into a collection of identifiable shards. As a result, deciphering a cascade of acronyms can occupy a researcher for their entire career. Gels fail, primers don't work, proteins misfold...but now I'm complaining again.

So how does this relate to fMRI? Well, I'm not entirely convinced that fMRI has earned its reductionist conclusions. Bloom does a commendable job criticizing fMRI for "turning bad explanations into satisfactory ones." But he never discusses some of the serious imperfections in the technology itself that are often glossed over. For example, in 2001, Professor Nikos Logothetis, of the Plank Institute in Germany, published a paper in Nature in which he simultaneously recorded the electrical signals of neurons and measured blood flow using fMRI. No one had ever done this before. Logothetis found that the increases in blood flow measured by fMRI do not necessarily parallel increased neural firing rates. In fact, increased blood flow can also parallel a constant, or even a decreasing neural firing rate. In 2004, Logothetis' lab found something even stranger . Neurons that had been chemically silenced - they could no longer become active - could still generate an fMRI signal that appeared active. As Logothetis notes, "This dissociation [between multi-unit activity and local field potential] could be observed in about 25% of the responses."

It gets worse. A 2002 study by Robert Harrison at the University of Toronto showed that fMRI signals "emanated only from areas endowed with a rich vascular network, and [that] no signals were obtained from adjacent regions in which the vasculature was less dense." Unfortunately, Harrison also discovered that in many cases the density of blood vessels in our brain has little, if any, relationship to our neural activity. Furthermore, blood also moves slower than the electricity in our neurons, so it's always difficult for fMRI to decipher what thought process the blood flow actually correlates with. Finally, whole parts of our brain remain invisible to fMRI machines. The base of our frontal lobe - a brain area crucial for consciousness - is too close to our nasal ducts to be visualized. The magnetism of air interferes.

To be sure, Logothetis and others think fMRI remains useful for measuring many different mental events, especially as imaging technology continues to improve. Even if our blood doesn't always correlate with neural activity, it can still be used as a pretty good proxy for neural processing. But we should not treat these anatomical portraits as definitive proof of anything. Pictures have a way of seeming more durable than most other types of data. The irony is that the opposite is probably true. fMRI isn't a window into the brain; it's just a crude map of our anatomy. The important thing is to not confuse the map of a place for the place itself.

Update 12/22/07: Comments have been closed due to an attack of spam.

More like this

This is a great intro to the debate. Thanks. I'm just moving into the field of MRI in rodents and we've had a number of people interested in fMRI. Looks like I have more reading to do.

It's hard not to love the images produced by fMRI. It's true that the science field (not to mention the media) has been seduced by them but they still have value.

My favorite quote about the failability of MRI (wish I knew where I read it) goes something like:

"Using MRI to discover the workings of the brain is like trying to fix your car from images taken by a geostationary satellite"

I think the papers linking a certain genotype to fMRI results can be pretty interesting and informative. I spent a post-baccalaureate year at the NIMH doing nothing special, but while I was there, some people in my group showed that a mutation in COMT, an enzyme involved in dopamine synthesis, has a noticeable effect on brain function:

"We then examined the effect of COMT genotype on prefrontal physiology during a working memory task in three separate subgroups (n = 11-16) assayed with functional MRI. Met allele load consistently predicted a more efficient physiological response in prefrontal cortex. Finally, in a family-based association analysis of 104 trios, we found a significant increase in transmission of the Val allele to the schizophrenic offspring. These data suggest that the COMT Val allele, because it increases prefrontal dopamine catabolism, impairs prefrontal cognition and physiology, and by this mechanism slightly increases risk for schizophrenia."

http://www.pnas.org/cgi/content/full/98/12/6917

Funny you should get into this. In an earlier post, you mentioned work by Samuel McClure's, which is nearly entirely fMRI-based. I attended his (1 hour long) thesis talk at Baylor College of Medicine, which consisted of 50 minutes explaining the signal of fMRI followed by 2-3 data slides (which actually consisted of 1 set of data shown 2-3 different ways). Of course, the fMRI explanation was mostly about the physics involved, since the actual signal being measured is largely speculation...

I was a little bitter, which only got worse when he and the neuroimaging group published the god-awful Coke vs Pepsi paper. None of this is to say that there isn't the possibility for good fMRI-based science, I just haven't seen much of it yet. Most of it is repeating interesting psychology experiments and looking for a differential fMRI signal. Given that they don't know what they're measuring, I don't think it deserves to be called neuroscience, maybe "anatomical psychology". Sorry. Still a little bitter.

By Crusty Dem (not verified) on 28 Jun 2006 #permalink

Jonah could you clarify something for me. You said:

In 2004, Logothetis' lab found something even stranger . Neurons that had been chemically silenced - they could no longer become active - could still generate an fMRI signal that appeared active.

I wanted to know what inhibitor they were using to chemically silence the neurons, but I can't find it anywhere in the paper you linked. Also, that paper was written in 2003. Could you update it so that you could tell me where you found that data?

This is a great discussion to raise. Certainly many ignore the extremes of data analysis that have to be adopted to produce those sexy pictures and leap off into ill-founded and long-winded discussions about what those pseudo-colored images reveal about the human psyche.

On the other hand, Logothetis's work didn't say that there was no correlation with neuronal signalling, but rather that there was a much stronger correlation between fMRI and "local field potentials" than with firing rates. LFPs are low frequency signals that are observed with a variety of intra-cerebral electrodes, and are thought mostly to reflect synaptic activity rather than spiking activity.

Thus, Logothetis's work implies that the fMRI signal provides a reasonable measure of the net neural input into a particular zone of the brain, which is arguably related in some significant way with how information processing is distributed in the central nervous system.

Well, I am going to be a "lab rat" for a study that is trying to examine what differences there may be, anatomically and functionally, between an autistic brain and a "neurotypical" brain. I am one of the autistic "rats."

I am doing this in hopes of being able to contribute in some way, however little that may be, to finding out the root causes of autism. I am not a scientist or an MD, just a guy with Asperger's syndrome looking for some help and hoping to contribute a little.

Part of this study involves sticking my head in the coil and seeing what my brain looks like and what parts lights up as I am asked questions, shown images, and so on. I will be but one entry of many into the data set of this study. I, personally, am very curious to see the results and plan to ask the researchers if I can be kept abreast of the results and given access to my data after the fact.

My son is also Asberger's and one of my nephews is mid-high functioning autistic. Autism seems to run in the family. I have a vested interest in this study.

I hope that fMRI studies are more scientifically valuable for this type of research than what some of you appear to be indicating. Reading this thread and the editorial seems to cast somewhat of a doubt on fMRI as a research tool for this sort of thing. I too, as just a member of the general public, have gotten the message that these fMRI machines are one step short of being a "true window into the workings of the mind." To be sure, that is marketing hype but, boy, I sure hope there is some value in these big, expensive hunks of advanced physics.

Thanks for the great comments. To respond to visualfx, there is tremendous value in much fMRI research. In fact, a recent study out of UCLA recently used fMRI to make an important connection between autism and mirror neurons. But I do think there are some technical problems that we still don't understand. Here's Christof Koch on the issue:
"It is generally assumed that hemodynamic activity is directly proportional to spiking activity. Thus, the larger the recorder fMRI signal, the higher the firing rate of the underlying neurons...Unfortunately, the relationship between these two isn't always so simple. Vigorous hemodynamic activity can go hand in hand with a decreasing neuronal firing rate (Mathiesen, 1998; Logothetis, 2001). Increases in the blood flow and oxygenation levels are most strongly coupled to synaptic activity, te release and uptake of neurotransmitter and the restoration of metabolic gradients, but much more weakly to spiking activity. The metabolic demand of generating and propogating action potentials accounts for only a small fraction of the total energy demand of the brain. Thus, the fMRI signal may primarily reflect synaptic input to a region and local processing, rather than neuronal output - the trains of action potentials that are sent to more distant sites." That's from "The Quest for Consciousness," p. 135

Also, per James comment: I apologize for the incorrect link. The relevant information is from a very insightful Annual Review of Physiology article by Logothetis. Here is the relevant paragraph:

Finally, exploiting the differential effect of neuromodulators on different cellular sites may experimentally induce LFP-MUA dissociation. In preliminary results in macaques from the Logothetis laboratory, neuromodulators were injected during simultaneous electrophysiological and BOLD neuroimaging experiments. Using a triple pipette (electrode, saline, and a neuromodulator), 20 microliters of 0.01 mol 5HT (5-hydroxytryptoamine hydrochloride) were injected over a period of 10 min. Several minutes after the injection a profound suppression of the MUA was observed. The LFP signal showed a slight increase and returned to baseline within a few minutes. During this time, in which the MUA was silenced, no significant change was discernible in the BOLD response. Spectrograms obtained before and after the 5HT injection during visual stimulation confirmed that the stimulus-induced spikes were entirely eliminated, although LFP activity was moderately increased. These measurements show that it is possible to dissociate pharmacologically spiking activity and hemodynamic responses. On the basis of the several dissociations described in this section, we conclude that the LFP signal is the key variable for the BOLD response.

Thanks Jonah for clarifying that.

I understand now. Basically they are using serotonin (5-hydroxytryptamine) to silence the spiking activity (MUA) but leaving the synaptic voltages (LFP) intact. It is not silencing the cells per say but does definitely reiterate your point that fMRI does NOT measure neurons spiking.