I had a paper accepted yesterday! This is my first first-authored paper, which means I saw this thing through from start to finish. And lived to tell the tale. I can hardly believe it myself.
This project began when I was a wee first-year graduate student some years ago (I will tell you flat-out that I am 30 years of age, loud and proud, but etiquette dictates that you NEVER ask a scientist how old their data are). I had just spent two years at the National Institute of Mental Health (NIMH) learning to scan brains, and I hoped to earn a spot in Rich Ivry’s lab at Berkeley by showing off my new skills. So when Rich, who would eventually become my PhD advisor, told me that he had a project in mind for me, I said great, I can totally do that in a ten-week lab rotation (HAHAHAHAHA I was so young and stupid).
The brain scanning technique I use is called functional magnetic resonance imaging, fMRI for short. fMRI is used to create pretty pictures of the brain in action, “lighting up” to reveal hotbeds of activity. But it’s not tracking the activity of brain cells, or neurons. It’s tracking blood flow, which is and isn’t a good proxy for neuronal activity, and I’ll tell you why.
The scanner is essentially a giant magnet, and the pretty pictures are made possible by the iron in your blood. Recently-active neurons receive fresh shipments of oxygen bound by hemoglobin, and the hemoglobin (heme = iron) changes its conformation (and thus, its magnetism) depending on whether it’s carrying oxygen or not. Follow the blood, the thinking goes, and it will lead you to active neurons.
Except when it doesn’t. Unfortunately, your blood is pumped up by your heart, which has a pesky tendency to beat faster or slower in response to THE SAME KINDS OF STUFF PEOPLE ARE TRYING TO STUDY. Scary pictures, math problems, ethical dilemmas, and even small movements all cause your heart rate to go way up or way down, so if you’re trying to learn how the brain responds to these things by looking for subtle changes in blood flow, well, godspeed to you.
Before there were fancy, expensive brain scanners, psychophysiologists in labs would hook people up to heart rate monitors, measure pupil dilation, monitor the small changes in sweatiness known as the galvanic skin response, and track the rate and depth of breaths, all to get clues about what’s going on in your head. This is, for instance, the basis of the polygraph, or lie detector test: the name just means “many graphs” (I assume), and by the way there’s a great book on the weirdo who conned everyone into thinking this was a good idea. I mean, in a way, it was: it’s a hell of a lot cheaper than fMRI-based lie detection and just as crappy.
I’ve said it before and I’ll say it again: Don’t bother studying the brain, the heart tells you everything. For example, when you’re anticipating something, it slows down just the right amount to allow more blood to build up–it’s thought that this happens so that when the time comes, you get a bigger pump of blood through your body. If you randomize the timing of events so that they can’t be anticipated, the heart learns the average and slows down that much. This gives me the creeps, but it also means that, like Peter Pan trying to dissociate from his shadow, researchers will have a hell of a time telling the difference between brain activity and “brain activity.”
I came into this problem where a fellow grad student and extremely wise mentor, John Schlerf, left off. John had previously had a dark night of the soul when, expecting to find a big fat “error” signal in the brain’s error processing center, the cerebellum, he instead found zilch. That is, until he noticed that errors caused the heart to more or less literally skip a beat, effectively canceling out that fresh supply of blood he was counting on detecting. So, using statistics and magic, he “corrected for” this change in heart rate and boom. Beautiful error signal, just where he knew it would be.
When I joined the lab, John and Rich wanted to do something very principled: to take a step back and ask how pervasive a problem this was likely to be. If errors could be masked in this way, what about other kinds of brain processing? My job was to study the brain’s response to simple, stripped-down arm movements. Poignantly, this kind of simple arm movement is what was used, in the early days of fMRI, to create a sort of template response to help predict brain activity. This template, known as the hemodynamic response function, can be thought of as a description of a person suspected of a crime. Say you believe a brain area is involved in some process, like movement, or memory, or reasoning. That area should give its location away every time that process occurs by “fitting the description.” Neurons fire, and the fMRI signal, known as the blood oxygen level dependent (or BOLD) signal, should go up in this sluggish, wavelike way. And wherever you see this happen in the brain, you color-code the area and say it “lit up.”
But what if that’s all just a bunch of blood being pumped up by the heart? Rich and John suspected that, if you removed the parts of the BOLD signal that fit a different description, based on recorded heartbeats, you might be left with, well, nothing at all, in a worst case scenario. This would have meant that all of fMRI was in serious trouble. And let’s be honest: I sort of selfishly wanted to be the person who showed this and published it more than I wanted all of the fMRI studies that had ever been done to be “real.” Don’t worry: it’s not all that bad, and I didn’t get to be the supervillain, the Hemodynamic Angel of Death, after all, as we shall soon see.
At the time, I knew this project would be a good way for me to become acquainted with the brain imaging community at Berkeley, learning to use a new scanner and new software packages. Because this was a methods project, I’d even have to dig deep into the guts of my code, hacking away at software written by, let’s be real, a total madman (I won’t name names but those in the fMRI community will feel my pain as I cursed myself for not sticking with AFNI, an NIMH-based package). This was wildly intimidating to me, but I knew I’d learn a lot and feel basically just real butch about my science. The goal wasn’t to figure out how the brain works but rather to figure out how we can best figure out how the brain works. It was not what I’d come for: I just wanted to make pretty pictures. But I also knew, from my time at NIMH, that methods projects were important, and to ignore these kinds of issues as an fMRI researcher is to consign oneself to reading very expensive tea leaves.
So! What did we do, and what did we learn? Well, first, we had people make some simple arm movements in the scanner. The rule was: Every time the crosshair turns green, you move.
We recorded their heart rate and breathing while they were in the scanner. Note that heart rate isn’t something you have a lot of control over, where breathing sort of is. You tend not to think about your breathing, but when we averaged together everyone’s breathing data, some people were rock-steady while others were more erratic, and so the effects this had on the BOLD signal were kind of a mixed bag. Heart rate, on the other hand, reliably soared after each movement:
This graph should scare you, because it looks so very much like the thing we’re trying to detect: the hemodynamic response, mentioned earlier.
We looked at two regions of interest, or ROIs, in the brain: the primary motor cortex (also known as M1) and the cerebellum.
See that nice, clean edge on the cerebellum ROI? It stops right before spilling out into the visual cortex above it, and that’s no accident. That’s months of hand-editing, a task I later outsourced to my undergrad minions, hoping it eventually took on roughly the same meditative quality as a mandala. Sidebar: I just recently learned that by going in and zapping blood vessels and other misidentified chunks of tissue, we were becoming intimately acquainted with the very same distinction (vessel or tissue?) that had, a decade earlier, caused Ben Carson to botch a high-profile separation of conjoined twins. Such a rich and storied legacy, that.
ANYWAY. Using statistics and magic, you take the files that mark every time your participant moved, you look ahead in time by creating a series of lagged files, and you pull out the BOLD signal from your regions of interest at each of those times to make a graph that, hopefully, looks like the canonical HRF and is a faithful representation of what happens in motor areas when you move.
Phew. Looks a lot like what we expected. So far so good. Then, you say WAIT, there’s ALL THIS OTHER CRAP WE RECORDED, like heart rate and respiration and whether the movements came just before or just after a heartbeat or breath and it’s all here! Let’s just throw that in and see what happens.
Once your statistics and magic account for all that other crap, guess what. It’s not THAT different. fMRI is saved. You can all go home, and go back to fighting with crappy code and bashing your heads against your keyboards.
Now, this isn’t the whole story, or the most recent version of it (these images were taken from old talks, because I’m not totally sure if I’m allowed to plagiarize my own figures before they’re even published. So, final results may vary slightly, but not much). And no, you shouldn’t really give up on the study of the brain just because it’s cheaper to study the heart (although there were times when I felt I should, and in my scannerless future, it’s definitely an appealing notion). All this says is that, on our hunts for brain activation, using the current description of the suspect should work out OK.
But then we do painstakingly show that monitoring and correcting for changes in heart rate and respiration, the way we did, can really clean up your data. We did a bunch of other stuff and made some really hideously complicated flowcharts that show exactly how much good each of our statistical corrections did–definitely worth looking into if you plan on scanning any brains attached to hearts. EVEN THOUGH we didn’t prove that fMRI is a sham and it’s all just heart rate getting in the way.
Shout out to the help and patience of John, Rich, and also Ben and Rick at the Brain Imaging Center (Ben’s blog, PractiCal fMRI, is fantastic–truly, he is doing God’s Work for fMRI researchers everywhere, and Rick heroically fixed our extremely expensive robot arm after I BROKE IT in a highly traumatic incident I was sure would cost me my spot in the lab. Extra shout-out to John for breaking the news to Rich for me, and to Rich, for taking me in anyway). And even though this took me YEARS (yes, if you’ve read this far, you get to know at least part of the secret) from data collection to publication (in my defense, this was not my only project), I’m so glad my baby is out in the world now. I didn’t know what I was doing when I started, but I do now. I mean, as much as anyone does.