OHBM2022 Keynote Interview with Anastasia Yendiki: On track with Anastasia
PhD student at the Max Planck Institute for Human Cognitive and Brain Sciences
Anastasia Yendiki is an Associate Professor in Radiology at Harvard Medical School and Associate Investigator at Massachusetts General Hospital, Martinos Center for Biomedical Imaging. Her research in the field of diffusion MRI tractography methods development started quite unexpectedly after she accidentally saw a tractography image at the Martinos Center on the day that she interviewed for a postdoc in functional MRI. Today, she is a leading researcher in anatomically-informed tractography methods development. In this interview, she tells us about what she enjoys in tractography research most… and also beyond!
Fakhereh Attar (FA). Your research is focused on tractography methods development. But it all started quite unexpectedly! Could you tell us what happened?
Anastasia Yendiki (AY). Yes, so I came from a background in tomographic image reconstruction. When I visited the Martinos Center for the first time to interview for a postdoc, everything about MRI and the brain was new to me. Then I saw a tractography image, and it just blew my mind when I learnt that you could follow the motion of water molecules in the brain and from there, you could reconstruct the arrangement of the fibers in the brain! It was almost like science fiction to me!
I found the physics of it really fascinating; I thought the mathematics of it was very interesting, and I instantly started to think about ways in which I could use my background in tomographic reconstruction in this new problem. Last but not least, tractography also came with striking visuals, and that was also something that I was drawn to. So, it was really a combination of all those things.
FA. How did your research develop into what you are focusing on today in your lab? What do you enjoy most about your current research? Why do you think it is important?
AY. I focused on reconstructing anatomically defined pathways for a while. So, we had fibre tracts of interest that had been defined in the anatomical literature, and we wanted to develop ways to virtually reconstruct those same fibre tracts automatically and robustly in a large cohort of subjects in vivo. It turns out that we now have good methods to reconstruct these anatomically defined fibre pathways, and tractography can do really well in this task. However, there are issues when you do not know the ground truth and you try to discover patterns of connections in the brain using tractography.
Currently, we are exploring how different types of ex vivo microscopy could inform diffusion tractography and help with some of its issues. What I really enjoy about this problem is thinking about creative ways in which we can bridge the large gap between the microscopic and the mesoscopic scales. I think that there is a lot of uncharted territory and a great deal that we can still learn about the connectional anatomy of the brain.
Bridging the gap between the microscopic scale of optical imaging and the mesoscopic scale of diffusion MRI is an important problem to solve, because we cannot take advantage of all the latest advances in machine learning without having the ground-truth connections in the brain to learn from. At the same time, even if we have this ground truth, it will be so vast that we cannot simply annotate it all by hand. So, we will also have to find clever ways to get around that. And I think it is a fun problem to think about!
FA. What different imaging modalities are you using in your lab to create the anatomical ground truth for accurate and robust tractography?
AY. We have gone down a number of different paths that have complementary strengths and limitations. For example, we use anatomical tracer injections in non-human primates. This technique gives you the complete trajectory of the axons starting from a certain injection site. As tracer injection studies are not applicable to human subjects, we have to figure out ways to translate the information across species. We have also been pursuing polarization-sensitive optical coherence tomography. We can apply this technique directly in human samples, but the field-of-view we can achieve is currently limited. That's something my collaborators in the optics group at the Martinos Center are working to solve at the moment.
FA. What are the most critical challenges you face in your research?
AY. First of all, the good news is that some of the challenges that the tractography community has focused on in the past have been addressed quite well. For example, it turns out that with today's advanced acquisition and analysis methods, we have gotten much better at resolving crossing fibers. So, now we have to focus on everything in the brain that is not a crossing. This is because, based on our postmortem validation studies, these are the fibre configurations that are causing problems—and that we haven't been paying as much attention to.
I find this to be an important challenge where ex vivo data can be helpful; first, to characterize the problem thoroughly and second, to find ways to fix it. I think that the microscopy data can help us better formulate the problem statement, which can then help us find a better solution.
FA. How do you think we can bridge the gap between the ground-truth microscopic information obtained ex vivo and maps of the fibre pathways obtained on mesoscopic scale in vivo? What are the factors that limit or challenge the translation of connectional anatomy ex vivo and in vivo?
AY. We need to be careful about having a representative range of geometries in the data that we are using as our ground truth: in other words, our training data. There are a finite number of geometries that could explain how fibers get from point A to point B, given the constraints posed by the shape of the brain and its various structures. Fibre pathways generally follow smooth trajectories, but there are areas where they have to take a sharp turn or branch or otherwise deviate from their main trajectory. That is where the characterization part comes in. Once we identify the problem—that is, the configurations that are causing the majority of errors—then we can focus on those problematic geometries. Maybe there will always be outliers (meaning some rare fibre configurations that we are not accounting for), but if we account for a handful of the most common ones, that will make a big difference compared to where we are now.
FA. Do we have to solve this problem separately for the long-range white matter fiber pathways, the short-range white matter fibre pathways and the intra-cortical fibers?
AY. The answer is yes and no! There are some similarities between the issues that influence the reconstruction of long- and short-range fibre bundles with tractography. One example that I mentioned earlier is the sharp turns. You can see sharp turns in the superficial fiber bundles near the cortex, but you also see sharp turns in the white matter. For example, fibers that travel from the frontal cortex turn down into the temporal lobe through the uncinate fasciculus. So, despite there being differences in the geometries of those fibres in different areas in the brain, there are also some commonalities in terms of the geometries that are confounding us at the moment.
FA. How would advances in diffusion MRI acquisition impact tractography methods development? Do we need to revisit the performance of our state-of-the-art tractography as new, cutting-edge diffusion MRI is acquired?
AY. Absolutely, I think that as the spatial resolution of the DWI changes, the considerations and confounding factors will also change. First of all, as we increase the resolution, we start to see more and more bundles that are real that we could not see before. But we also start to see streamlines that are not real. Increasing the spatial resolution will thus increase both true positives and false positives. So, we have to revisit where we are along that trade-off. Think of it as an ROC curve. As we change these acquisition parameters, we move along that curve.
FA. Tractography is in wide use today. What are the common errors you see tractography users make? Do you think we can use tractography to explore new white matter structures?
AY. The problem is thinking that you can use tractography to replace anatomy; i.e., when tractography is used to discover patterns in brain connections that have not, and in some cases cannot, be corroborated by other techniques. By “other techniques,” I mean ones that can image axons more directly, without relying on water diffusion.
If there is no corroborating evidence from any other technique, then it is very dangerous to assume that you have discovered a connection just based on diffusion tractography. This is because of the limited spatial resolution and also the indirect nature of the measurements. So, exploration and discovery is fine, but at some point we must confirm those findings with a different approach.
Tractography is not at a point where it can be our only source of information on whether a certain connection exists or not. I'm hopeful that some of the work combining diffusion MRI with microscopy might help us move in that direction one day.
FA. We’ve covered the challenges of tractography to some extent. But tractography isn’t just challenges and imperfections. How do you think tractography as a tool has assisted us to advance our knowledge of the whiter matter connectional anatomy?
AY. I think that one of the great benefits of tractography was that it spurred a lot of new interest in white matter anatomy. Tractography enabled us to segment white matter structures that can’t be resolved with other in vivo structural imaging modalities, where the white matter appears uniform. In doing so, it created a lot of interest in going back to the anatomy work of the last decades (and even centuries); in some cases, rediscovering what anatomists had previously discovered with their own techniques. I think that we should definitely continue going back to the discoveries made with the “slow research” approach of anatomy and use them to better understand the discoveries we’re making now with the accelerated research that’s possible with tractography.
FA. Where do you see tractography going from here? What is your favourite research currently conducted by other labs?
AY. Ultimately, we want tractography to help us identify brain circuits that are implicated in disease, so that we can then intervene by perturbing those brain circuits—hopefully, towards healthier brain function. When we ask where tractography needs to go from here, and how accurate it has to become, it’s helpful to answer that in the context of how reliably we can use it in such a task.
So, I can aspire to see every single axon in the brain. But why? We can set very ambitious goals, but if we do not know why we are setting them, they might not be the best use of our resources. Ideally, the goals should be set by the developers and the end users of tractography together.
In terms of research conducted by other labs, there are several efforts to use microstructural indices to inform tractography or vice versa, which is a very interesting idea. This is another area where microscopy can help answer open questions. For example, these techniques assume that microstructural indices are continuous along a bundle but different between different bundles, and that diffusion MRI is sensitive enough to capture those differences. These assumptions can be investigated with post mortem microscopy. We know that they probably hold for certain special cases; for example, motor fibers that are more heavily myelinated and thicker than the association fibers that they cross with. But how generalizable is that to the rest of the brain? I think that there are some fascinating questions there to pursue.
FA. Now, Let us step away from tractography and talk about your career path. What were the challenges and opportunities you experienced along the way?
AY. In terms of the challenges, I guess I’ve had to learn to deal with being underestimated. Because it is something that has happened a lot: it still happens, and I think it will just continue to happen. Earlier in my career it was very frustrating because I felt that being underestimated was something that closed doors or opportunities, so it kept me from advancing. But once you get to a point where you’ve survived long enough despite it—and you've found ways to get around it—then it might actually become an advantage. Because it gives you the element of surprise! So it is something that is not going to go away, but now it can be a bit more entertaining to me than it used to be.
As far as opportunities go, the key opportunities in your career are those times when you find people that you work well with and that you can build meaningful collaborations with. They can be from any career stage: more senior, peers, or more junior. The common thread is shared values and standards. You don't need a lot of them, a handful of meaningful collaborations is enough to make a difference in your career. A big part of feeling at a good place right now is because I'm lucky to have that.
FA. In that light, would you have any advice for the young researchers who are considering pursuing a scientific career?
AY. That’s an interesting question. As much as I try, I really cannot come up with any advice that applies to and is helpful for everyone. So perhaps the advice is exactly that: if you hear someone say that they found “the” way that we should all be doing things, be very skeptical.
FA. Finally, what do you do beyond research that you really enjoy?
AY. I have been learning to dance flamenco for well over a decade now—with the emphasis on “learning” because it is a lifelong process! Even when you get to the point where you might be able to teach it, as I have, you are still learning! In some ways it is like science, because it is very much about the relentless pursuit of precision. But it's also about self-expression. It is like being the musician and the instrument at the same time!
More recently, I have also started to learn the flamenco guitar, which is even harder! I'm just getting down to the point where I'm not terrible at it! So, still a lot of learning for me to do on all fronts, both scientific and non-scientific.
Thank you very much Anastasia for the interview, and we look forward to your keynote lecture at OHBM 2022!