Dr. Peter Bandettini — Mr. MRI
Dr. Peter Bandettini spends a lot of time peering into people's heads. Not because he is clairvoyant, but because he is a biophysicist. Using functional MRI (fMRI), a revolutionary neuroimaging technique he helped pioneer in the '90s, Dr. Bandettini delves into the mysteries of the human brain. He is working to advance fMRI technology to parse out more information about the neural connections that are constantly and spontaneously active even when we think our minds are blank.
Dr. Bandettini is a principal investigator at the National Institute of Mental Health (NIMH) and the director of the Functional Magnetic Resonance Imaging Core Facility. Learn more about his work at https://irp.nih.gov/pi/peter-bandettini
[Strange mechanical noise]
>> Diego (narration): For anyone who’s never had the pleasure of hearing that beeping and thumping before, that’s the sound of an MRI machine.
Having recently gone through my first MRI, I’ll just say the experience can feel a little… bizarre. First, you have to strip down and remove all metal from your body—the machine uses a very, very powerful magnet. An MRI often has a magnetic field of 3 tesla. A typical refrigerator magnet is approximately 0.001 tesla. Meaning that it would take around 3,000 refrigerator magnets to equal the magnetic field strength of an MRI machine. At that point, it’s more the fridge sticking to the magnet and not the other way around.
So, after you triple check that you didn’t leave on any stray metal for fear that you’ll go flying across the room, you’re greenlit for entry. You lay down on the patients table, which then slowly rolls into the narrow center of what can only be described as a car-sized doughnut. Once inside the giant metal pastry, you have to keep perfectly still, while the clicks and clanks seem warp time so that 20 minutes feels like an hour. It’s all pretty trippy.
But of course, MRI isn’t just a claustrophobic’s worst nightmare, it’s a revolutionary tool used in medicine and research. MRI stands for Magnetic Resonance Imaging and it consists of the strong magnet, radio frequency coils, and a scanner that together open a window inside the human body.
Doctors use MRI to visualizes detailed images of the organs and tissues and help diagnose certain conditions or diseases, such as tumors, aneurysms, or stoke. Fortunately for me, it was just my lower back that needed the once-over.
In the lab, MRI can be used to measure and map brain activity over time—a technique called functional MRI, or fMRI. Dr. Peter Bandettini was one of the early pioneers of fMRI, having helped develop the technology itself.
Dr. Bandettini is a biophysicist, principal investigator, and director of the Functional Magnetic Resonance Imaging Core Facility at the National Institute of Mental Health where he has been using MRI to dive into the mysteries of the human brain for decades. He has written a book about the fundamentals of MRI, aptly named fMRI. And most recently he is the host of “Brain Experts” a podcast all about neuroscience.
I had the pleasure of Zooming down with Dr. Bandettini to pick his brain about ours.
>> Diego (interview): Thank you so much for taking the time out today.
>> Dr. Bandettini: Sure
>> Diego (interview): Before we get into like the specifics, can you give me a quick rundown of how MRI works and what exactly it is that you see with MRI?
>> Dr. Bandettini: So, MRI is a technology that's been around for a long time. It turns out that certain nuclei have a net magnetic moment such that they resonate. And lucky thing of nature, that water which makes up, you know, most of our bodies is one of those. And you give it an RF pulse or radiofrequency pulse at that frequency, that excites or that gives energy to the system. And then in a very short time, about 50 milliseconds to 100 milliseconds, it gives it back.
>> Diego (interview): So, it’s kind of like reflects?
>> Dr. Bandettini: It's more like an echo. It's more like if you yell in a canyon and then a short time later, your echo comes back. It's kind of like that. And actually, we use the term echo a lot in MRI. So, basically you give the RF power and the signal echos back, and then while it's echoing back, you can apply ways of getting the spatial information. So, you can actually make images of where all the water is in the brain as it's giving energy back. And that ability to map was invented in late '70s by Paul Lauterbur, and he actually got the Nobel Prize for that. And then by the '80s, really quickly, that was adopted clinically. And so, it was immediately really useful because it was the only technique that could actually image soft tissue the best. It was better than CT or X-ray or things like that. So the brain lent itself perfectly to that.
>> Diego (interview): Right. And what's so revolutionary about MRI is that it doesn't use like harmful radiation, right? That's like a huge game changer in the medical field.
>> Dr. Bandettini: Yes, exactly. Exactly. It's all -- it's all RF power. So it's like putting, you know, a radio antenna up to your head. There's nothing really invasive. The only thing -- Typically the only thing that people worry about with MRI is the acoustic noise and it's usually about 90 to 110 decibels and so people have to wear earplugs and that's about it. But less than 10 years after it came out, another discovery was made. And I was lucky enough to be in one of the departments as a graduate student in 1991, developing technology for this, but there's a way of doing really high speed imaging, and it's called echo planar imaging, in which you collect a whole plane of data in one echo. So usually it takes multiple RF pulses, this one -- one RF pulse, you get a plane. So, it's super fast. You can collect the image in 50 milliseconds. And you can make a movie, you can make a time series of these movies. This was the first time in about 1991 where just a few groups were trying this fast technique to make movies and then having people tap their fingers or shine light in their eyes and see what happens and they're able to see brain activation occurring. And the reason why they can see this is because blood has a really unique magnetic property: it's paramagnetic, all other tissue is diamagnetic. What that means is that when you apply magnetic field to it, it weakens the magnetic field slightly. But with blood, if it's unbound to oxygen, it actually concentrates the field slightly.
>> Diego (interview): Does that have to do at all with the iron in our blood?
>> Dr. Bandettini: That's exactly right. It's due to the iron in the blood. It turns out the iron when it's bound to oxygen, it's diamagnetic. When it's unbound to oxygen, it's paramagnetic. And that slightly distorts the magnetic fields around the red blood cells. So when you actually image a brain, you see all the vessels and they’re dark. All the blood is like black. And then it turns out that if you tap your fingers, what happens is in that area that's active, oxygenated blood rushes in, and there's actually an overabundance of oxygen that's in the blood in that area all of the sudden. It takes about 10 seconds to sort of increase all the way to the plateau.
>> Diego (interview): Yeah. Yeah.
>> Dr. Bandettini: And because of that overabundance of oxygenation, just a little bit, that allows us to see what parts of the brain are active.
>> Diego (interview): Gotcha.
>> Dr. Bandettini: So we're able to be very sensitive to blood oxygenation, as it changes dynamically in the brain with activation.
>> Diego (interview): Right. And like you were saying, so MRI isn't just like, you know, static images. It tracks brain function as a function of time. So, what kind of studies use MRI?
>> Dr. Bandettini: So there's two types of MRI that I like to differentiate. There's MRI, which is sort of the broadest one. And then functional MRI is sort of a subspecialty within MRI that mostly researchers do. But one big aspect of fMRI that people have been developing over the last 28 years has been designing experiments in which the behavior is extremely well controlled or well understood or the stimulus is very specific such that, you know, they're really able to delve into and pick apart exactly what part of the brain is responsible to specific types of behavior. And they've been doing that for a long time.
Just to mention, there's another type of functional MRI that has come around, also 20 years ago, but it only started catching on about 15 to 10 years ago, and that's called resting state fMRI. And that's actually kind of has caused a revolution in fMRI. And the idea is this, is that even if you're lying in the scanner, doing absolutely nothing, it turns out that your brain is always sort of churning. Even if you're not consciously thinking of any thoughts or specific, you know, you're not having specific tasks.
>> Diego (interview): Yeah.
>> Dr. Bandettini: It's constantly spontaneously active. And turns out that the areas that are spontaneously active that show fluctuations are in synchrony. And so, we now have developed techniques that can map out all the parts of the brain that show synchronized spontaneous activity. And it's a beautiful map. We could actually map out up to 120 different regions that line up pretty precisely if we were to do specific tasks that try to tease them apart. So, you know, we get motor cortex, we get visual cortex, we can break the visual cortex down into all its hierarchical organization. All using just simply resting state; just putting a person in the scanner and having them do absolutely nothing.
>> Diego (interview): And do we know what that like steady state of activity means in the larger, human context? Is it just kind of like the routine things that our body does that we don't have control over?
>> Dr. Bandettini: That's the mystery. People are still trying to figure out why is it that we have this spontaneous activity that's so robust? So, people think, OK, it could be keeping a certain tone of activity such that you can respond rapidly to the environment or whatever. Some people think that it is conscious sort of mind wandering. Part of it could be subconscious. Also, other parts, people argue that it's pure vascular physiology that, you know, it's just sort of oscillating, and that areas, you know, fire and, you know, they fire together. And like you said, it could be micro movements that we do when we're lying there, whatever, at least in the motor cortex. I think it's a little bit of everything in some sense.
I have a postdoc who, just before she came to my lab, she wrote this beautiful paper—her name is Emily Finn—characterizing this resting state activity. And turns out that there's specific networks in the brain that show more robust activity as a function of, she measured, their intelligence. And so, as a function of intelligence, they all sorted out in terms of how much this network was spontaneously activated. And so, we’re kind of at that stage where we're kind of just poking and probing and trying to figure out what correlates with what. And that's another challenge of fMRI is getting down to characterizing individual subjects—like, you know, based on intelligence, based on, you know, she also just wrote a paper in my group on trait paranoia. And this was a little bit different task. This wasn't a resting state test. They were just listening to a story. It was sort of what's called a naturalistic stimuli, where they're listening to a story. As they were listening to the story, they were looking at the oscillations in each subject. And to the extent that they correlated with each other, across subject, you know, it was a time-lock sort of story. So they're -- we kept the timing the same.
>> Diego (interview): Yeah.
>> Dr. Bandettini: It turns out that that the subjects who were more paranoid, generally had more paranoid characteristics, showed more correlation in certain areas and the ones who were less paranoid showed more correlation in other areas. And so you're able to pull out trait paranoia.
>> Diego (interview): Why were you looking at paranoia as a factor?
>> Dr. Bandettini: I think that the rationale—if I think back on that, there's nothing particular about paranoia that we care about, really. I mean, it was more just –we’re trying to come up with easily measurable traits. So, you know, you could have impulsivity, people have done that. Intelligence is the obvious one to test. You know, people looked at reading ability in our group as well. And paranoia was just another one of those things. And so we were trying to come up with easily measurable behavioral traits and probing to see if there are neural correlates to this that can differentiate the individuals. And we're finding them. It's surprising how well fMRI works with this regard.
But now the question is, you know, what does it mean? What do these brain areas actually indicate? It could be an epiphenomena. It could be something fundamental. And that's one of the criticisms actually of fMRI is that we're still at the stage where we're just sort of, you know, we're collecting data and trying to make sense of it. And then we're slowly trying to build models of, you know, network models of the brain to try to figure out OK, what's actually going on here?
>> Diego (interview): Well, speaking of that, most recently you’ve used the technology to look into working memory. So, do you want to walk me through those experiments and what you found?
>> Dr. Bandettini: So, working memory essentially is, what you have to use all the time if you read a telephone number and then have to remember long enough to dial it in. It’s just sort of like what you hold on to just for 20 seconds. And so it turns out that there's a part in the brain that seems like it's always activated with working memory. And that is what's called the dorsolateral prefrontal cortex. It's kind of like right behind your left eye -- between your left eye and your ear -- in your left ear.
>> Diego (interview): OK. So, specifically to the left.
>> Dr. Bandettini: Sometimes the right is indicated, but usually it's the left dorsolateral prefrontal cortex.
>> Diego (interview): OK.
>> Dr. Bandettini: It turns out the amount of activation in the dorsal lateral prefrontal cortex predicts pretty accurately whether you will remember something or not. And so we could then predict whether someone would remember a sequence or not based on the amount of activation in their dorsolateral prefrontal cortex. So the question was, well, what is it that the dorsolateral prefrontal cortex is doing? And one theory is that it's sort of a hub communicating with the other parts of the brain that process this information to sort of keep it in play, in some sense. And if it is doing that, so if it's communicating out to other areas, then there's some idea that the upper layers that receive signal in from the other areas are active.
Unfortunately, like a lot of scientific experiments, we didn't necessarily answer that question, but we did find some interesting results. We found that, during working memory, it wasn't the whole dorsolateral prefrontal cortex that was active. It was just the upper layers, just the upper part. The cortex is sort of like this, you know, this sheet that's all crumpled up in the brain. And the sheet has a thickness that's about five millimeters thick. And it turns out that the upper few millimeters of that cortex are what are selectively active during the maintenance phase of working memory. Then the subject had to, at the end of that period, they had to respond with a button press, depending on whatever the task was. And it turns out that when they responded, sure their motor cortex became active, but what's interesting is that at the moment that they responded, the lower layers of the dorsolateral prefrontal cortex became active. So, essentially what we think is happening is that the memory information is consolidated in the upper layers, but then the lower layers send out the signal to the motor cortex. So, then suddenly we finally have somewhat of a circuit diagram of being able to differentiate, OK, the upper layers are active, and then suddenly the lower layers are active that send out the signal to actually make the response.
So, you know, relative to invasive electrophysiology, it's still somewhat primitive, but this is a, I think a big jump in fMRI, to be able to differentiate input versus output activity. Then we can potentially take that information and start building network models of the brain. All the network models are based are nodes that are acting on other nodes. And so now we can say, oh, this node is acting on this node based on the fact that the activation was in the lower layers that send signal out.
>> Diego (interview): And you can only see those lower layers because of the higher intensity magnet?
>> Dr. Bandettini: Yes, exactly. The higher field-strength magnets, they give us a lot more signal to noise to work with. So that if we tried imaging this at 3 tesla, we could image at that resolution, but it would be way too noisy. The functional signal would be way in the noise. So now at 7 tesla, it's popping out of the noise, and we can actually see it.
>> Diego (interview): Oh, OK, that makes sense. So it's not a matter of like—you were able to see the same picture but now you can just see more of the activity?
>> Dr. Bandettini: Exactly. So typically, on a 3 tesla, the highest resolution people would go to is like 2 millimeters.
>> Diego (interview): Gotcha.
>> Dr. Bandettini: Whereas with 7 tesla now they can go to 0.7 millimeters with the same signal to noise as at 2 millimeters at 3 tesla. It’s like building a, you know, a bigger telescope that can collect more light in some sense where you can get finer and finer images.
>> Diego (interview):That’s funny, I was going to ask because I recently had like my first MRI and these things are huge. They’re the size of cars and they're just as noisy with all the clanks and the beeps and all that stuff. And I saw a picture of the 7 tesla getting like craned into the building. So what does the future look like? You were saying that it's almost like building a bigger microscope, but we're not trying to make them bigger, right? Like, I'm assuming we're trying to get smaller machines or like what's -- where is the technology heading, in your opinion?
>> Dr. Bandettini: Yeah. No, there's different directions the technology is headed. One direction that magnets are going is yeah, to make them smaller and reduce their footprint. You know, people are trying to build 3 tesla scanners that are just for the head. And so you can put them in any sort of hospital room that's not specially equipped to put a magnet in. You can sort of like just plunk it in there. It's not too heavy. And actually, there's even another company that has come out. One of my former students is in this company that makes these super low field scanners. These really, really low field scanners that don't have superconducting magnets, just have standard magnets that you can wheel around and put into an emergency room or put into an ambulance or something. So, that's sort of a technology that's very, very, very new.
>> Diego (interview): What can you see with that? Is that good enough to see like—I mean, I would assume that it’s not as powerful.
>> Dr. Bandettini: Yeah, it's not as powerful. You can barely—the images are really bad. But they're good enough to sort of triage. I mean, they're good enough to sort of put someone in and say, oh, you're bleeding, or something like that. It's good enough for a doctor to make a decision if they have to make a decision quickly.
>> Diego (interview): And you said you could use them in like an ICU or something like that? It doesn't have that potential danger or you know how like the magnet strength, obviously, you're supposed to strip down and have no like metal on you. You can still do that with these portable machines?
>> Dr. Bandettini: Yeah. It's a low enough field strength such that it's not even that dangerous to have it—it only is dangerous maybe like within a couple of inches of the scanner, as opposed to the big scanners that we have where the whole room has a magnetic field and you have to be very aware. So that's another direction that it’s going. But maybe the most important direction is—so right now, there’s this limit we're up against with our 11.7 tesla scanner and that is, the scanners that are superconducting, they need to be chilled to an extremely low temperature. And the only way to do that is with liquid helium. And there's not an unlimited supply of liquid helium in the world.
>> Diego (interview): Right. Isn’t there like a scarcity right now?
>> Dr. Bandettini: Yeah, and it's worrisome because you can't just manufacture it. And one of the things that's been holding us up from ramping up our 11.7 tesla scanners that we can't find helium. There's no market that were even able to buy the helium.
>> Diego (interview): Right.
>> Dr. Bandettini: If we buy it, it's super expensive. So now a technology that companies are ramping up is scanners that can be chilled using a very, very small amount of helium that doesn't boil off or doesn't leak or whatever. So that's -- that technology is just, you know, just try to keep MRI going without needing liquid helium.
>> Diego (interview): Does the helium get used up as you do more scans?
>> Dr. Bandettini: So it doesn't get used up with the scanning. Eventually it just diffuses right through and leaks. It's hard to find a containment that for helium in which it doesn't leak out of. Like back in the '80s, we had to replace our helium every few months. We'd have to come in with these big dewars because it would just leak out. It would boil off.
>> Diego (interview): Right, and it’s not like you can go to like Party City or something and just get a tank of helium.
>> Dr. Bandettini: Exactly. Exactly. So they're getting much better at containing it—of only needing a small amount. I think most scanners now you don't need to replace the helium for a long time.
>> Diego (interview): Well, I don't want to neglect asking you about like your personal story and how you got attracted to working with MRI, and no pun intended.
>> Dr. Bandettini: Yeah. I've always been interested in brain function. I think that, you know, the brain is the most complex thing that we know of. And it's also one of the biggest mysteries. Even now, we have no idea what a neural computation is. And so I gravitated towards that. I was thinking of going to medical school. I decided not to and I started graduate school because I wanted to—I liked physics but I also like neuroscience and I want to try to combine them somehow. And I got really lucky. I went to—it’s not a very big graduate school, it's a medical college in Wisconsin—but it was a great place because they worked on MR hardware and so that allowed them to always be at the bleeding edge of what's possible to do with a magnet. And I naturally wanted to image the brain and image brain function. And I was working with a colleague, Eric Wong, who built these gradient coils out of sewer pipe and wire that you would stick in the scanner to allow it to go fast. And just at the time, there were some inklings of functional MRI being possible, so we quickly programmed up the sequences and got that going and we collected, with my head, motor activation. And we were like the third group to do it, but we were lucky enough to be the first group to actually publish a functional MRI paper. And so ever since then I've been completely fascinated with pulling out more information from the fMRI signal.
I mean MRI is really a unique technology in a sense. It's not like X-ray. I mean, X-ray is useful in some sense or other imaging techniques. But MRI has so many degrees of freedom in which you can explore information. So you can tweak something and pull out different aspects of physiology. Like you can pull out diffusion. You can pull out flow. You can pull out connectivity. And now with fMRI, you can pull out, you know, the fMRI signal changes and the dynamics and now the fluctuations. And so, ever since then, I've been working on sort of this interface of applications of fMRI and method development. So we’re trying to methods that might lend themselves to applications that might derive information about the human brain.
And so, yeah, so I finished my PhD. I did my postdoc at Mass General Hospital and Harvard Medical School and came briefly back to the Medical College of Wisconsin. Then started here at the NIH in 1999. And started up the Functional MRI Core Facility and my research section and we've been growing ever since. And now we have, you know, four 3 tesla scanners, one 7 tesla scanner. And in the last few years, because of the way the field has been progressing, I've started up another group, a machine learning group, sort of basically using machine learning to further, not only process the data, but sort of, to better model the data. To actually have real workable models of brain function based on the fMRI data.
>> Diego (interview): And, what will it mean to have like this complete model of brain activity? Is there a point you think where we’ll understand, I don't know how many years in the future, but we will understand everything about the brain, or are there things that you think there won’t be answers to?
>> Dr. Bandettini: Yeah, that's a good question. That's a really good question. So I like to use the analogy of Darwin, you know, came up with natural selection. Before, where biology was a little bit was sort of just documenting things. And, you know, you characterize the diversity of nature. And it took a certain sort of a leap of insight to get a principle of what's going on. So I think we're kind of at that stage where, you know, we're really characterizing the brain at many different spatial scales. And, you know, some people claim that, oh, it's the cellular scale that really matters. Or understanding the genetic influence, which it might be. Or right now, with fMRI, it's sort of cartography in a sense where we're characterizing all the areas. And let's say we understand exactly how they're connected and communicating, we still, we still need to understand more of the principles of brain function. And the only way of getting at that, I think there's a lot of potential in sort of the parallel tracks of like AI versus neuroscience. You know, AI is trying to come up with better intelligent algorithms. And in some sense, it's using neuroscience to try to come up with principles of having better AI systems. So, I think we are still nowhere near there in terms of neuroscience to understand what computations are done and what are the principles of how the brain is organized—how it develops and how it evolved. So we have no principles, right now, we have tons of data.
>> Diego (interview): Yeah. That’s a good way to put it.
>> Dr. Bandettini: Yeah. So we're still at that stage right before, you know, someone has to come up with the equivalent of natural selection. And hopefully fMRI can play a role in that.
>> Diego (interview): OK. Well, lastly, I do want to ask a question that you ask your guests on your podcast. What advice would you give to someone who's interested or maybe like in the early stages of their career in biophysics or, you know, neuroscience?
>> Dr. Bandettini: Oh, I would try to give people advice to always try to step out of their comfort zone, especially when you're younger. You sort of get tunnel vision, and you learn all the jargon, you learn all the concepts and you sort of hone in. And it's really interesting. And that's good. But the real progress is made by people who synthesize ideas across disciplines and across spatial and temporal scales. Trying to do that as much as possible, trying to see how your work fits in to the bigger picture on a regular basis I think will not only help the field, but it will also I think, help people's careers in a sense that they would never reach necessarily dead ends. You know, a lot of times the field sort of becomes so rarefied and so specific that that it seems like it's interesting, it's penetrating, but you have to also connect it. And also, especially in fMRI, right now my advice, a more practical advice: it's become a lot more computational. So, most people who come in my lab, sort of a requirement is that they have to know Python, and they have to know how to process and they have to be comfortable working with data and working with models and that sort of thing. So, that's a big thing, but at the same time do that but don't lose sight of the fact that, you know, nature and the brain are really complicated and it always helps to sort of ask questions that are sort of try to bring together other disciplines and are outside the box.
>> Diego (interview): It's good advice, practical and bigger picture. Well thank you again, so much, for taking the time.
>> Dr. Bandettini: Ok, well thank you, thank you.
Sound effects courtesy of freesound.org