Talking with Professor of Biomedical Engineering Michael Miller

Read story

Interviewer: What have you done in your academic career that you believe to be creative or innovative?

Michael: Over the past two decades, I’ve pioneered a field called computational anatomy, which attempts to create systems of understanding for medical images. It works something like this: Say we could take a 3D magnetic resonance image and go from the order of one billion measurements to a declarative statement like, “This person has a myocardial infarction,” or “This individual is converting from normal to mild cognitive impairment,” or “This individual has temporal lobe aphasia.” Being able to have a machine do that is quite challenging. We don’t have any machines that can literally go from the many, many measurements to knowledge representations. That really is the holy grail of image understanding.

Now our greatest image understanding systems are artists and painters. You can think about what Seurat was doing when he created pointillism. You start with the canvas, and the canvas has thousands and thousands of points. There’s no notion of a knowledge representation. There’s no notion of the house in the point or of the person in the point. That’s a deeper construct. You know that Noam Chomsky really explored this very deeply with computational linguistics. Chomsky forced us to look at this; he had many examples of this. The notion of—he called it the surface structure—which would be the points in the pointillist image, or in Chomsky’s case, the phonemes of the words in the sentence, and the deeper structures, which are the knowledge representations, and building the deeper structures. So Chomsky forced us to realize this sort of interconnection between the surface and the deep knowledge representations. Sentences like “time flies like an arrow,” and “fruit flies like bananas,” forced us to really look at the syntactic and semantic parts, the deep structure, the knowledge, and its relation to the surface, so the notion that form and function follow each other.

Picasso, of course, did this in his Man with a Pipe. There he really pushed us. Picasso was maybe the greatest image understander we’ve ever had. He totally shattered the surface. He made it contradictory, yet always there was the concept of the man smoking the pipe. So we’ve been trying to do this for shape, form, and function in medical images, and that’s what computational anatomy is.

To give you a sense of how that could work, or how that works in computational anatomy, think of global positioning systems. Global positioning systems are really at the heart of image understanding. They’re very much like what you’re doing when you’re parsing a sentence, separating it into its parts, its structural parts and its functional parts. In a global positioning system, you carry around your iPhone or your Android. If it’s the iPhone, the company that delivers that is Skyhook, which allows you to know where you are at all times in world coordinates. We in medical imaging call world coordinates the atlas, or the textbook—the knowledge representation. World coordinates are interesting because if you’re walking down the street and you want to know where the Starbucks is, you take out your Starbucks app and it delivers that to you; it’s positioned you relative to world coordinates. If you’re interested in getting a haircut, you go to Groupon, a company that has grown faster than any other company I think in history. Basically, Groupon delivers coupons to you of something you want in the location you’re at. All of that is global positioning.

We built essentially the technology for global positioning. The global positioning system that we built in computational anatomy is called Diffeomapping or Diffeomap. It basically builds correspondences between coordinate systems, and it’s based on map making. If you get out your American Automobile Association atlas or you get out your Michelin road map, that’s being done on every page. There is some notion of how one part of the map corresponds to another part of the map. In the case of computational anatomy, we take your personalized coordinate system, whether it’s your brain or your heart, and we find the correspondence between it and atlas, or world coordinates. Now why that’s interesting, as you can imagine, is if you now have world coordinates in correspondence with your personalized brain or your personalized heart, then we can bring to bear all the knowledge that we have of world coordinates in the atlas. This is really the basis of personalized medicine, and in the context of medical imaging, we call this “high throughput imaging informatics.” The informatics is essentially the knowledge layer, the meta-layer that you derive from the surface, the pixels, and the measurements. Man with a Pipe is the meta-layer: that’s the knowledge. In the case of the declarative statement, “This person is converting from normal to mild cognitive impairment,” that’s the meta-layer. And it might be associated with a structure in the image. That’s what we’re really trying to do in computational anatomy and why it’s interesting, because of imaging informatics.

Interviewer: When you say “world coordinates,” you don’t literally mean where someone is out in the world, but you’re comparing this specific instance with the known and previously recorded topography of populations?

Michael: That’s right. World coordinates will be that we have Gray’s Anatomy—it’s a continuum of atlas representations of, say, a human brain. Currently, all our pediatric patients are going through this process. In pediatric imaging and radiology—which is Sue Mori and Jon Lewin—for every neuroimage that we take there, we actually build this correspondence using Diffeomap. We can then layer on top of it this informatic layer, which is the world representation of all the associated diseases. There are 28 characterized diseases that we see in pediatrics kids that are all in spatial correspondence; diseases that are associated with anatomical structures, subcortical structures, gyral structures, and other parts of the body.

Interviewer: What I understand you’re telling me is that there is a differentiation between health and disease state, focusing on an abnormality that, through scanning process, you can identify and compare. So is there just one normal?

Michael: No, no. There’s a continuum of normals. In the world atlases, we literally have regressed against age and against disease. We have a continuum of knowledge representations, all spatially and physiologically located in coordinates of real biological tissue. It’s that information that we bring to bear, appropriate for gender, appropriate for disease state, appropriate for age type.

Interviewer: Certainly you just described a field that is remarkably creative and innovative. In terms of our own insights, did you have an aha moment or was there research that you had done and you said, “Wow, I just figured something out?”

Michael: So I understood that in the global positioning systems that we’ve had for a long time, there’s a very simple correspondence that’s being done, and it has something to do with the field of rigid bodies or rigid body mechanics. If you take an airplane, the only way you can change it is to rotate it or translate it—that is, affect its roll, pitch, or yaw or change its position. Rigid bodies—the theory of rigid bodies and the control of rigid bodies—is the basis of modern control of airplane flight and the basis of robotic control of actuators. Google can use it in Google Maps because there you are rigidly aligning geographical objects like you did at the board when you took a triangle and you said, “Euclid taught us that this triangle is congruent to this triangle”—what you did was you rotated it and translated it, and that was your notion of congruence.

Now what I appreciated was that, in fact, human shape and form is not rigid. It’s a deformable body. This notion of a finite dimensional correspondence, which corresponds to the typical global positioning system, is actually a six-dimensional correspondence, a rotation in three spaces each: three dimensions in a translation and three dimensions in rotation; that’s six dimensions. But that doesn’t work for deformable bodies or for human anatomy. We don’t all grow into the same shape and size. In fact, you can think about that transformation being locally distributed and when you go to look at this local distribution you end up in an infinite dimensional setting.

It turns out that the theory of deformable bodies we’ve constructed—which generalizes the theory of rigid bodies—is all about these kinds of very high dimensional transformations. We call them diffeomorphisms. Morphism means structure-preserving mapping, and diffeo- is a particular instantiation associated with the invertibility of the mapping. These diffeomorphisms are infinite dimensional versions of this very simple notion that is the basis of your global positioning system. While global positioning systems wouldn’t work for comparing you to me, these diffeomorphisms will work because you can think of them as very locally defined. In a place where your nose has to get bigger or smaller or your eyes have to get bigger or smaller, it allows you to do that while at the same time maintaining the morphism property, maintaining the structure that keeps tissue connected and keeps typology conserved. That was my contribution in the ’90s—to appreciate that. We could literally construct this theory of deformable bodies using notions from classical equations called the Euler-Lagrange equations in continuum mechanics, which correspond to flows of diffeomorphisms.

Interviewer: Was this something that you published in a paper?

Michael: I’ve had a series of papers. In 1997, with Ulf Grenander, I had a paper called “Computational Anatomy: An Emerging Discipline,” and it was really where we tried to do what Chomsky did for computational linguistics, something he did in three papers in the 1950s. We tried to build a generative theory that would allow you to start with an equation and be able to build anatomical structures that are as rich and complex as human anatomy. At the core is this notion of these morphisms, these diffeomorphisms, these flows of diffeomorphisms.

Interviewer: What I’m struck by, in this conversation, is that I’m hearing “Noam Chomsky,” “Picasso,” references from arts and culture. You’re not talking purely in terms of your own field. When you are trying to do something innovative, is that part of the whole process, being able to go out from your own discipline and pull in other ways of understanding and seeing the world?

Michael: Well, I like to say that biomedicine is sort of the Manhattan Project of medical solutions. If you think about what went on in the Manhattan Project, you can have a sense of the reason why biomedicine is coming to be called convergence science. Biological systems and biomedical systems occur across essentially 10 orders of magnitude in spatial scale and temporal scale. From the molecular level, the program project that I’m part of with Marilyn Albert as lead is a dementia project that comes out of the National Institutes of Health. It’s one of about 10 around the world. There are people working on molecules, and there are people working on integrative function of the brain at the functional one millimeter scale. When you work across these scales, these spatial and temporal scales, it’s really not going to work to stay within a narrow domain. The academy evolved—there are specialties in physics and biology and mechanical engineering and electrical engineering and civil engineering. However that’s not the way biological systems evolved. They evolved using all of it simultaneously.

Every problem that I’ve tried to solve has involved many different disciplines. I was very interested in building global structures, doing image understanding, and building global structures from images. As a graduate student, I was doing single-unit recording, and I couldn’t see how to go from single neurons to discovering how the brain understands things. But Chomsky was in the air, so I read Chomsky, and that’s exactly what he was doing. He was going from words to the noun phrase. The What question—what does the What question mean? It means, What was used to hit the nail? That’s called a What question. Well, you can see that that’s an integrative concept, and if we want to get to integrative constructs, if we want to go up in scales, we’re going to have to understand other disciplines because all these disciplines work on problems that have different aspects.

I’m in the world, and I see Picasso and I know, I’m certain, that if Picasso were alive he would get this. Picasso has a beautiful movie—have you seen this movie? He stands behind the canvas, and he clearly understands the grammar—literally, the grammar, the Chomsky grammar—of shape creation, which I don’t. Even though I have these rules of generation of anatomical shapes, it’s not as beautiful as what Picasso does in this movie. Of course he’s drinking his wine and he’s sitting in his chair behind the canvas. The photographer’s there, and Picasso puts one thing up and then he puts another thing up, and you see a fish. Then he does one more stroke: it’s the haiku of form. He does one more stroke, and you see a duck. Then he does three more strokes, and you see a house. And then you see a little girl walking to the house. It’s all of 11 strokes. You’re a thinking person, and you say, “Well, that’s what Chomsky’s generative grammars were. How am I going to understand the shape and structure and connection of the human brain? Perhaps there is something here.” So you go and you look.

Interviewer: Picasso as “image understander” as you call him is part of a broader revolution in how we understood our world thanks to Einstein. You often hear it said that the 20th century was the age of physics and that the 21st century is the age of biology. Do you concur with that and if so, why?

Michael: Well I concur, but it’s not quite right. The first half of the 20th century was the age of physics, and of course Harvard and Chicago bet the farm on that. MIT and Stanford bet the farm on the information age, from the work Claude Shannon had done. The Mathematical Theory of Communication by Shannon and Warren Weaver—a 1963 book—was a popularization of what Shannon had done in 1948 in a two-part article by the same title. So Shannon only defined what information was in 1948. The information age was created—and all that wealth associated with MIT and Stanford was created then. It was really associated with the information age.

Google ended that in a way because now we can search all that’s known almost instantaneously, and it’s likely that soon we won’t even have to hit a browser to do that. There will be a technology that will connect our thoughts to that browser. So we’re really in the post information age. We’ve really moved out. We’re no longer asking the question, “How can we get at all the information?” Now we’re asking, “What can we do with the information?”—which is where your question was going, I think. So, yes, I concur. Certainly Shannon wouldn’t have done what he did if quantum mechanics hadn’t happened. So they share it together. We’re moving into the computational biology age and this age of huge growth, just this astronomical growth, in data. As this all becomes available to all of us as well as our ability to search it, it’s changing the way science will be done. So, yes I agree.

Interviewer: What do you imagine the end result will be?

Michael: Well I don’t know what I imagine the end to be, but I know what we’re in the middle of. If you look at Wikipedia, for example you just have to know that as universities, we’re moving into a new era. The printing press was probably the start of the universities. I don’t know when the first university was, but this availability of information, the availability of all this information on the Cloud, is of course going to change everything as we know it. The flattening of the earth—what Thomas Friedman talks about—is that we now have billions of people who have access to this same level of information. So for instance they are going to be looking at their own outcomes in health care. They are interacting via social networks and creating all sorts of new interactions and new ways of being. I don’t know where it’s going to end—it’s not going to end.

Interviewer: So this is the new age of ubiquitous information and limitless collaboration? Is that what is happening in your discipline?

Michael: Biomedical engineering has its root at the interfaces of disciplines because of this notion of crossing scales. That’s why I would say it’s a convergent discipline. We don’t have a choice if we work on biological systems but to bring people from different disciplines together. One of the things I can say about myself, which seems to be perhaps different than many others, is that I don’t really think I am all that creative at solving puzzles or addressing long-standing problems that are well known in a field. But what I do seem to be good at is connecting dots.

In biomedical engineering, we are faced with problems that have many, many different aspects. When I think about this notion of collaboration, I think about two things: First, is the context in which you’re placed. When I was a graduate student, I came here and I was in the first-year medical school class because the first year of biomedical engineering is all medical school classes. I learned anatomy and physiology and biochemistry from people like Albert Lehninger, who is a giant in the field. We had four new Ph.D. students that year. The somatosensory homunculus was just being measured by the likes of Mike Merzenich, and Sol Snyder was just discovering opiate receptors. So there were the four of us in the seminar sitting in the front row of that very small room in the Traylor Building. The space was this size [measuring out with his hands]. These people were talking to me about the brain and the physiology of the brain, the structure of the brain. That could never have happened if we weren’t in a university.

At the same time, as I look at the rest of my career, I spent 15 years working with Ulf Grenander who is one of the greatest living applied mathematicians. I have had the opportunity to have a National Science Foundation award with David Mumford who’s a Fields Medalist in algebraic geometry and to work with Laurent Younes in the Department of Applied Mathematics and Statistics. This just wouldn’t be possible if we didn’t have a research university. There would be no context. I mean how could that happen any other way? I would say that collaboration and the notion of a research university in which you are supported to work on fundamental things and go figure out the solutions, wherever they are, those two things go together. For me, I can’t think of anything that I have done, any work that may be distinguished, that wasn’t collaborative, that hasn’t involved others helping me do it. It’s very clear that you have to collaborate because of the range of problems.

Interviewer: Do you think that connecting the dots that you talked about is a uniquely human capability?

Michael: Well, I think there are so many different kinds of creativity. People say to me that I am creative, but I think there are kinds of creativity that makes us all different. I am sure there is something integrative about my brain that doesn’t allow things to sit in a place and not get connected to another place. It just doesn’t work that way. On the other hand, in terms of the creativity of solving a very hard puzzle, that’s not always me. I don’t get interested in it and maybe I don’t get interested in it because I know I don’t solve them. So I think there are many different kinds of creativity and intelligence. You saw that with Picasso, right? I mean he seemed to go through, I don’t know, five phases, and he was doing different things in each phase. It was amazing actually. Picasso had a long career, and there were many different ways he was being creative. So I think connecting the dots is a good one, and I am lucky that I have it in my genes. I am fortunate to have it.

Interviewer: Is it something from childhood, from an early age?

Michael: I wasn’t conscious that way. I wasn’t thinking that way. I am very much a Long Island Jew who did great in all of his classes and played sports. I wasn’t differentiated in college, but then when I was a senior I thought biomedical engineering was great and I should go for that. I have been on this path, but I have never been that ahead of the game that I thought of myself in special ways. Now retrospectively, I think that I feel honored to have spent significant amounts of time at Brown as a visiting professor and significant amounts of time in Paris every year, and I have been at Washington University and I have been at Hopkins and I have just had a wonderful go of it. So I am lucky.

Interviewer: Are there insights that you have gained about the way the brain functions, the way that we think, and the way that we create, that perhaps others just wouldn’t know? That in the general population there wouldn’t be an understanding?

Michael: Well, I think yes for me, but yes for me and the other 500 neuroscientists who are part of the Brain Science Institute at Johns Hopkins, which is a magnificent place in brain science. This idea of the organic basis of who we are is something that those of us in science and especially those of us who have a neuroscience slant we really believe in. So the manifestation of my connecting the dots, I believe, is going to turn out to be a gene that has something to do with the reticular activating system or something that directs waves of energy across my brain that allows me to see different things in different areas and connect the dots. The taxicab driver builds up a very strong representation of England in his hippocampus and has a very detailed map—that discovery wasn’t surprising to me. I am not surprised to see that musicians have a particular part of their temporal lobe that is very well developed. I think that appreciating the organic basis of who we are—that it is a combination of our genotype, our phenotype, and our environment, all of which come together to create who we are—may be different than most people. We see a lot of who we are in the structure of our brain. I and the other 500 neuroscientists at Hopkins probably feel that way, but I am not sure that most people in society think that way.

Interviewer: Is that idea threatening to them?

Michael: Well, I think that it’s complicated because we all want...I have faith, I still want to have faith that there is something going on. I have a Buddhist brother, and so if this removes all the Gee whiz then it can be a little disconcerting. But I don’t necessarily get bogged down in that thought too much. It’s too much fun just exploring the science of it.

Interviewer: In newspapers an evergreen story that they’ll run all the time—and therefore must get a lot of readers—are stories about someone has a traumatic brain injury and suddenly they have some clear manifestation, say that they can remember faces but not names. It seems like, up until fairly recently, much of our understanding of the brain was this sort of coarse understanding. There’s the famous example of Phineas Gage in 1848 who had a railroad spike driven through his brain and survived, but became a different person so he was “no longer Gage.”

Michael: Yeah, yeah. But you asked me if functional imaging is a very coarse scale tool? Of course, if you’re thinking about neurons, then it’s a very coarse scale tool. But if you believe that much of function is laid out in anatomical structure then the examples that I think of are like these: your auditory system lays out a frequency representation in the cochlear nucleus and in the auditory system very logically, with the tone representation in your somatosensory homunculus. Your whole body is laid out logically. You touch a point of your body and it’s all laid out in the cortex. So if you believe that structure and function—the Chomsky paradigm—that they live together, then you conclude that if you can look at integrative scales on the order of a tenth of a millimeter to a millimeter (which allows you to look within gyri and within the mid-brain sub-cortical structures and look at white matter connections), you can see things happening at that level of detail. Then you also believe that you can essentially measure function and see the brain functioning so that this is an integrative aspect that goes from the single neuron to the patches of millions of neurons. I know that’s a far stretch—I was a single unit recording person in the auditory nerve for years. Now to be working at the one millimeter and the half millimeter and the tenth millimeter scale, is a little bit of a stretch. But I think we have very refined tools now—much more than the spike going through the head—so that we can watch you as Humphrey Bogart walks into the room and then Bergman walks into the room, and we can see how men react and we can see where they react, and we can see how women react. That’s highly informative. We’re seeing the brain thinking.

Interviewer: Is that an actual research project you’re talking about?

Michael: Yeah, well, I mean they do it at Kennedy Krieger. They have a wonderful set-up. There are many, many exciting things going on. Charles Limb is in Otolaryngology—he’s done rapping in the scanner. There’s just a load of beautiful, wonderful things, just looking at how we function in the world as beings. I think functional imaging, structural imaging, connected to genes and proteins and all of it taken together is allowing us to look more deeply at how the brain works.

Interviewer: Do you imagine that an enhanced physiological understanding of the functioning of the brain might permit us chemically or surgically to enhance brain functions someday?

Michael: Well absolutely. I mean we already have. Michael Merzenich was part of The Brain Fitness Program on PBS. Now Mike Merzenich was a student of Vernon Mountcastle, and I guess he graduated when I was just arriving. He’s a famous neuroscientist. He’s the one who’s created games and things for you to do to enhance brain function. These puzzles are all about this—they exercise your brain. But don’t you think that Michael Jordan being able to jump five feet high as a result of training and genes and all the rest, that the equivalent isn’t going to be true in thinking? Of course it is. So yes, I think that we have to figure out how to exercise the mind and keep it vibrant. Hopefully we can make that part of our educational system.

We have to figure out how to keep it healthy. Whether that will be give it pills—I’m sure there will be some of that—but hopefully we can actually drive it; use the mind to exercise the mind. Just like we run laps, and I go out on the treadmill every day. It wasn’t until recently that we realized that for years people who have their first myocardial infarction would then report depression. They’d be depressed. And we’d all say, “They’re depressed because they just had a myocardial infarction.” Well that’s true. But they’re also depressed because when you have this disease your heart is throwing up all these little things into your brain and in many cases you’re having little mini strokes, little micro strokes. Guy McKhann, leads a beautiful program project that has sort of shown over the years that brain disease is associated with heart disease. That tells you how primitive we still are. It just seems so obvious in retrospect, yet it’s only recently that we really understand that.

Interviewer: You say throwing things up – is that little electrical impulses?

Michael: No. Little plaques—the same thing in your heart that’s plaquing, that causes a blockage, is essentially throwing little micro plaques into the brain, and your cerebral vasculature is getting impacted, causing brain disease. I’m bringing that up just to say that we really don’t know much yet about how to really build healthy brains, how to exercise your brain.

Interviewer: There’s the brain as computer, as calculating machine. Maybe theoretically through exercise or drugs or better nutrition that may make everyone smarter. But the other key to it is judgment, compassion...

Michael: Danny Goleman wrote for the New York Times science section for years, and was the one who wrote the book Emotional Intelligence. It was the first one that popularized the notion of IQ and EQ, intelligence quotient and emotional quotient. That’s what I think you are referring to, so yes, there is an emotional quotient. They’re complementary. Now people are showing that emotional quotient it’s extremely important in terms of predicting success, perhaps as important as IQ. I don’t know how much we’re doing at Hopkins. There’s a lot of work being done now, certainly in California, on the happiness center, on being happy. That whole aspect of what it is that we do that makes us happy, and then how does it show up, what’s the organic basis of it in your brain? I’m not sure how much is going on at Hopkins, it’s sort of a functional in between we may not cover, but it’s important. It’s really going to be important because we know that depression is a significant aspect of mental health and having people be happy is a good thing.

Interviewer: You talked about how certain things could only happen in a research university just because of what it does. So proximity and bringing people together obviously is something that the university does that fosters innovation or fosters creativity because it allows this to happen. Can you think of other aspects of how it works, how the university functions to promote that?

Michael: Well, our university, Johns Hopkins, is a wonderful place. I was at Washington University for 15 years, which was a wonderful place of similar size and focus. Historically, Johns Hopkins has been very influential in defining the modern research system. The research system I’m referring to is this notion that faculty are essentially entrepreneurs. Faculty go and sell their programs to funding agencies, and it’s through funding that faculty are able to support their time. Of course, part of the formulation of the university is to support you a significant amount of time so that it makes it possible for you to do research. I think that model is extremely vibrant at Hopkins, although we’re all, of course, concerned about what’s happening on the government side.

What seems to be happening at Hopkins, which is maybe not unique, is this whole notion of the matrix of departments. You make departments rows, and you make centers and institutes columns, this notion of going across departments. If you think of a matrix or a center as a vertical, like a vertical market—in business, they call them vertical markets—then centers and institutes are like verticals, so it’s a particular market. Maybe you’re in the neuroscience market or the brain imaging market, or I have a center in imaging science. So we’re in the imaging science market. Then there’s a nano market and there’s the computational medicine market. Those tend to go across departments. That seems to be, over the last 10 years, a model that has worked extremely well. The departmental model still prevails, but departments are disjointed from each other, and faculty see mostly other faculty in their own departments, but they don’t tend to cross departments. But centers and institutes are really doing that.

It’s interesting to think about whether that’s really going to be the long-term reality and the model by which research will be supported, that research will naturally flow to. Historically, it was always centered in the departments completely. Now, these centers across departments are really hugely involved in the research missions. If you look at schools now, there’s a significant contribution by centers and institutes to the bottom line, in terms of research. That’s what funds research. I’m not sure if that’s where things are going, but the downside of that and the concern of the departments is that you could interpret it as the flow away from a departmentally focused model to a center and institute model. I think part of the discussion now at the university and at universities is related to this, because we’re working in convergent science. In convergent science, we need to bring people from different departments together. The departmental models don’t do that, but the centers and the institutes do.

Interviewer: How important do you think is physical proximity? Or is it?

Michael: I think physical proximity is everything.

Interviewer: Even in the days of the Internet?

Michael: It’s everything. Having said that I have a fabulous joint effort with Susumu Mori in radiology, it’s called the Center for Brain Imaging Science. It’s part of the Brain Science Institute, which is all over the campus. This is a major focus for me now. Yet Susumu’s group is in one place and my group is another place. But I think that works because we’ve gotten to a scale. Maybe this is true for companies. When companies get to a scale where they can project, they don’t have to be co-located. But I couldn’t have done this 10 years ago. Ten years ago when I moved here, I really needed to have a group around me that I was working with, so that I could formulate a core that I could possibly instantiate and then support somebody else’s activity with. So I tend to think that being together is a good thing. Co-location is very important and we need to figure that out. I certainly think new scientific discoveries occur because people are on the same hall. They see each other and they interact.

Interviewer: Crick and Watson, their understanding of DNA came sort of from having been in contact with people who were doing something different. That strikes me as sort of typical.

Michael: Well, in my own work I couldn’t do it if I didn’t do it with applied mathematicians, with people in computational biology, and people at the medical school in radiology who do imaging. I just couldn’t do it. It wouldn’t be possible because the scale is too big. So this idea of working on these convergent themes is key—they cross boundaries and it’s challenging.

Interviewer: Where do students fit into all of this, in terms of their presence and in terms of your interaction with them?

Michael: Well, I’m on the Homewood side, and so I’m closer with undergraduates. I love my students. Even though I’ve spent my whole life doing research and being a highly supported investigator, I love my students, my undergraduates and my graduate students. And I tend to have fewer postdocs. On the medical school side, there are really fewer undergraduates. Undergraduates go down there, but undergraduates are the center of the campus here, and on the medical school side I think there tends to be more postdocs. I have a strong graduate student program, and my graduate students tend to be the ones that publish and go on and take faculty positions, and really work on my research topics. So computational anatomy has been created by the 10 of us, me and three or four colleagues and then the 15 or 20 students that we’ve graduated and that’s who’s out there doing it now and that’s why it’s legitimate. Then it catches on, it grows.

But on the undergraduate level, we have a major resource at Hopkins in biomedical engineering. We have almost half of the undergraduate engineering student body, and those 100 students are amazing. Every one of them loves the program. They come to my class; I teach a core class in systems and control. In biomedical engineering there were only four core classes. We built this curriculum when I got here. I saw this at MIT. At MIT in electrical engineering, every undergraduate student takes four classes. They are the four core classes, so they have a common electrical engineering experience. Then they go out and they say, “I’m starting a company. I’m leading your company because I’m an MIT electrical engineer.” They’re branded. They have a view of what it is. It’s those four core classes, because they’re all in it so they get to meet all their other kids and that’s their network. I saw the power of that. You’re creating a field when you do that, right? So at Hopkins, we have four core classes, and all of our undergraduate students take the four core classes. And I teach one of them. I mean, it’s a hard class. And they all know it’s hard and all the students before them tell them it’s hard. But they love the class and they love the faculty because we care for them. We want them to succeed. We want them to learn. But they’re the ones that are going to go and make the world, because all start-up companies and all ventures, it’s all the undergraduates who do it.

At MIT, the donors they track are all their undergraduate students. When they went and built their incredible building, they had a list of all their undergraduate students. And those are the ones that—10 percent of them were the donors that built that building. The undergraduate class at Yale or the undergraduate class at Princeton it’s the same thing—we often associate with where we were as undergraduates later. So I would just say that in Hopkins’ future, we have to turn loose this intellectual energy of the undergraduate class in a much greater way. We have this magnificent medical school that’s huge—it’s amazing, if you look at its size, if you look at the research engine it’s like $500 million. We have to figure out how to increase the role of undergraduates so that they can play the same role in all of the other departments that they play in biomedical engineering. They’re amazing.

Interviewer: Do you have thoughts on how we can figure that out?

Michael: We have to get this right. This is the future of biotechnology. The bio-information economy is going to be significant. You know that the Cloud is growing at an unbelievable rate. A friend of mine at MIT estimated if you look at the growth of Flickr, by the year 2020—that’s 10 years—there are going to be 10 to the 20th images in the Cloud. To give you a scale of what 10 to the 20th is, I call it the “human click limit”—if you take every person on the planet and you have them take a picture every second for 100 years, that’s what 10 to the 20th is. IBM suggests that currently 30 percent of the Cloud is medical images and medical information. So you’re going to have this huge biomedical information economy. I’m not the first one to have this thought. All the companies are appreciating this, and in China they are appreciating this. But the interesting thing is that the major medical schools—the universities—will be the stewards of these information clouds, because they will be creating the information. So that opportunity is quite an opportunity. And we really have to understand that opportunity. I think Hopkins will. It’s really a major opportunity. So we’ve got to get right this understanding that we’re moving into a new era. Look at Wikipedia: we in the university, we’re not trained to think about what happened with Wikipedia—this idea that you have this crowd, a billion people, all contributing to the creation of knowledge. There’s something called the Open Source Software Project. NIH adopted it. MIT started it in the ‘90s. But now there’s this open classroom idea. And Stanford just had a class with 140,000 participants. The first Artificial Intelligence class ever that was open to everyone. And it was just weeks ago that it started. So one gets the feeling that something’s coming and it’s moving quickly. Recently, Thomas Friedman was on “Meet the Press” and he pointed out that when he wrote The World is Flat in 2005 he didn’t mention Facebook in it. It’s all happened—and happening—very quickly. And so we have an opportunity. We’re training the thought leaders, the knowledge experts, what Chomsky used to call in linguistics, “the linguistics knowledge experts,” the people who understood language, who’d write the grammars down. We’re training them. We understand what’s in the data. We need to be building applications that run off of these Clouds. So this is a great opportunity for us. But we need to get it right. We need to figure it out.

Interviewer: Being ready and getting it right brings up your discovery of computational anatomy. Was the world ready for that? I mean when you sort of started doing this, was everyone like, “Oh good, I’m glad. We’ve been waiting for someone to do this!”

Michael: No. No. No. You have to understand that...I say this to all my students all the time, “When you write a paper that you think is an important paper, if it doesn’t get completely destroyed in review, it’s not an interesting paper.” That’s just the way it is. When we first wrote “Computational Anatomy,” we submitted it to a particular journal. I knew the editor. The reviews that came back were scandalous. But that’s fine. There were other journals. There were lots of people around who appreciated. We were asked by a whole bunch of different ancillary places to write little magazine articles about what we were doing. So we knew it would get published. But in general, I’ve never had a lack of confidence. I mean I may at times have a lack of ability, but I’ve never had a problem with confidence. And you just have to believe in what you’re doing. You have to do what you know is cool and you believe in. And then it always turns out. But no, the world wasn’t ready. But within five years of two or three of those articles, now every single person in medical imaging uses the word diffeomorphism. Nobody knew what it was. Everybody used the same mathematical model, the same equations of motion. It’s all a question of the way that we’ve delivered them and implemented them numerically. Now it’s a question of, “Are there more efficient ways to do it?” Other groups are doing that. But everybody is working in that framework that is the framework for understanding human anatomy and medical imaging. And now we have to keep moving. We have to now link these structural and functional phenotypes down to the genotype, which is going to be its own thing. We don’t do this because somebody else says it’s a good idea. If somebody else says it’s a good idea, it means it’s already out there. It’s like the stock market. It’s already all in—it’s already factored in—by the time you hear anything, the price is already in.

Interviewer: So one of the constants throughout human history seems to be resistance to the new idea. Why?

Michael: Why? Because I’m trained to believe. What I always say is that I completely believe what I believe until you show me I’m wrong. That means we’re strong willed in that we’ve built a structure that’s based on assumptions and proofs. It’s not going to be easily touched, but it gets shifted around. It’s part of the process, the head-banging process. I’m a good scientist, and what that means is, that as soon as you show me I’m wrong, I then believe that. I move forward with that.

Interviewer: Good. Anything else I should have asked you?

Michael: No. No. It was great.

Interviewer: Well, thank you very much.