Why fMRI is unsatisfying – a neuronal perspective

461889a-f1.2Each year when the Society for Neuroscience Meeting rolls around, all of the major journals devote extra space to neuroscience, publishing hot articles to attract the attention of the 30,000 plus attendees at the conference.  This year is no exception, and one of the most important articles came out this past week in Nature with the heady title “Intracellular dynamics of hippocampal place cells during virtual navigation“.  The paper, by Chris Harvey, Forrest Collman, Daniel Dombeck & Dave Tank is a tour de force investigation which combines new technology with insightful experimental manipulations and shows, according to an accompanying commentary by Doug Nitz, that “it is not impossible to examine brain correlates of higher cognitive processes and at the same time identify their underlying causes at the cellular level”.

The detailed results are probably too technically specific for most people in the field of neuroethics, but this study highlights some of the reasons that hard-core neuroscientists view fMRI with disdain.  Given the prominence that imaging the human brain has come to play in neuroethical discourse, I encourage readers to take a few moments to at least try to appreciate what the issues might be.

First, let’s take a look at what Dave Tank’s group at Princeton have done.  For over 35 years, neuroscientists have known that the firing rate of a subset of hippocampal pyramidal cells (the so-called place cells) change in predictable fashion as the animals navigate through a spatial environment.  In particular, the firing rate of a place cell reflects both the animal’s present spatial position and the path the animal has taken to reach that position.  Think about that for a second: the output of a single neuron reflects a highly nuanced and information rich algorithm.  But it does not stop there.  When multiple place cells are recorded at the same time, they exhibit a phenomenon called phase precession.  Nitz’ commentary sums it up nicely:

The firing order for a set of hippocampal place cells with partially overlapping place fields is found to match the animal’s physical trajectory corresponding to those fields. Phase precession stands as perhaps the most robust example of temporal coding of information in the mammalian brain.

So we have a nuanced algorithm which operates at both the single cell level, and in even more remarkable fashion, across groups of neurones in the hippocampus.  What has eluded neuroscientists until now is how the synaptic information is integrated to produce this phenomenon.  Enter Dave Tank’s group who developed a technique whereby they could carry out intracellular recordings of place cells while the animals approximated natural motion by running on a large ball whose movement was immediately translated into the visual projection on the screen by the open source video game Quake 2.  You can see a very cool demonstration in the video at the bottom of the post, but I want to return to Nitz’ effusive commentary:

The broader promise of the technique lies in learning exactly how the myriad incoming synaptic potentials to any given neuron are integrated to yield spike-firing patterns that closely track specific thoughts, perceptions or actions.

Let us now return to the issue of fMRI.  I don’t want to rehash old arguments here about the problems with spatial and temporal resolution of fMRI, as they are probably known to most readers.  What I do want to draw your attention to is the overall objective of fMRI: to be able to visualize the brain in action, and to derive from that information some insight into how the living brain does what it does.  In the mouse experiments, the key observation was that phase precession was encoded by small changes in the membrane potential of place cells, and that these changes arose secondarily to synaptic inputs. In other words, the experiments provide an initial glimpse (and I really mean glimpse – these data are fantastic but they only hint at the kinds of remarkable insights that will come in the future) at what appear to be the real workings of a complex cognitive construct – encoding not just the location of the animal in space but the trajectory by which it arrived there – and this phenomenon is manifest at the subcellular level.

I must say that I empathize with the despair that the smart and well-intentioned people who have put a great deal of honest work into developing thoughtful fMRI protocols should might feel upon reading about these new data.  For I think it is inevitable that this experimental result raises a new round of substantive questions about whether the BOLD signal can provide the type of insight that fMRI practitioners seek.  My conclusion is that, barring some major technological advance, it does not.


2 thoughts on “Why fMRI is unsatisfying – a neuronal perspective

  1. The study you are discussing is quite interesting, but it’s sad that you decided to title the post and segway it towards a complete nonsequitur critique of fMRI. Every technology had limits and strengths.
    If anything, from what I’ve seen is more and more scientists are burying the hatchet about fMRI. It has it’s uses and many lower level researchers are using the results to inform and guide their work. You’ll never directly image phase unlocking with fMRI, but other methods can’t get whole brain localized metabolism/blood flow measures either. If some method could do everything, that’s what we’d be using.

Comments are closed.