Brain-Machine Interfaces, Deep Learning, and Neural Modulation

Share with your network

How biomedical neuroengineering can make our collective future brighter 

Eva Dyer’s research centers on machine learning and neuroscience, and developing computational methods that uncover principles governing the organization and structure of the brain. Chethan Pandarinath’s research focuses on neuroengineering, deep learning, brain-machine Interfaces, and neural coding, as applied to devices that assist people with disabilities and neurological disorders.

Dyer and Pandarinath, both assistant professors in Emory and Georgia Tech’s Coulter Department of Biomedical Engineering, were awarded 2019 Sloan Research Fellowships honoring early career scholars, “with a unique potential to make substantial contributions to their field.”

We spoke with them about their research, the future of biomedical engineering, and what brains and birds have in common.

Both of your research lies at the intersection of biomedical engineering and neuroscience. What is the big question about the brain right now?

Chethan Pandarinath

CP: My area of research is neuroengineering, which focuses on repairing the nervous system in cases of injury or disease and also on developing new tools or quantitative approaches to study the brain and further expand our understanding. We bring in cutting-edge computational techniques from computer science and artificial intelligence to better understand the brain, and then try to use these insights to develop new therapeutic approaches to treat brain injury or disease. There’s no question to me—this combination of new tools is going to fundamentally change how we understand the complex processes in the brain that make us who we are, and if we can start to compare these processes between healthy individuals and cases of injury and disease, we can develop new strategies for repairing the brain.

ED: One of the major questions that is driving my research is a question of heterogeneity and variability. In particular, when we study disease models, we would like to learn signatures of disease from data. Because brains are so different in their structure, however, they all produce different measurements. I am fascinated by how we can build expressive models that capture this variability to build better decisions about brain state and disease. This trend toward personalization in health care could end up being critical in the study of brain disease as well. With all of these topics and projects, engineering is critical because we need to design new data analysis tools that enable discovery in large and complex datasets.

You each have a background in electrical engineering and now work in the BME field. What do you find exciting about BME right now and where do you see BME heading in the future?  

CP: I think this is an unbelievably exciting time for our field. For a long time, neuroscience has been really limited by a lack of tools. The brain is incredibly complex, made up of billions of individual neurons that are intricately connected into powerful networks. Yet, in terms of neuroscientific experiments, we’ve mostly been able to study the activity of one or a handful of neurons at a time. When you’re staring at the activity of one neuron, it’s extremely challenging to say anything concrete about what a billion-neuron network is really doing. There is a confluence of factors now that promise to transform our understanding of the brain. On the one hand, we’re getting new tools that allow us to monitor many thousands of neurons simultaneously, and these capabilities are growing exponentially. When you can monitor the activity of 10,000 neurons for long periods of time, you start to create truly massive datasets, which will hopefully allow us to ask fundamentally different questions about how neural networks in the brain really work. At the same time, our ability to process complicated data has transformed over the last 6-7 years, thanks to the rise of a field of artificial intelligence known as “deep learning.” Now we can build artificial neural networks that are capable of processing large and complex datasets in very sophisticated ways, allowing us to uncover relationships in data that we’d never be able to pull out in the past.

Eva Dyer

ED: As we continue to see exponential growth in data analysis and the computational sciences, it has become increasingly attractive to bring these tools to bear on problems in biomedical research. BME is becoming increasingly dependent on the use of data for discovery – and this trend seems to only be growing in BME and other areas. BME is really about combining engineering and biology, and thus, as we continue to see problems arise that require that integration of engineering and biology, biomedical engineers will be at the forefront of many of these integrative innovations. As the size of datasets that are being generated in neuroscience continue to grow, it will be even more critical that we use machine learning and data-driven methods to help us uncover the mysteries of the brain. As we scale up our learning methods, we will be able to apply the lessons learned from simple controlled experimental tasks and scale these ideas to more realistic and naturalistic settings which will not place constraints on behavior. Being able to record and interpret the activity of neural populations “in the wild” will likely contribute to major breakthroughs in our understanding of the brain.

Where do you see your own research headed?

ED: A lot of the projects in my lab now focus on learning variability in large-scale neural datasets—this could be individual differences, changes in brain structure and function due to disease, or even changes that occur during learning or development. The aim is to develop automated systems that can navigate through large datasets and identifying changes that could be due to any of these factors. With automated approaches to discover changes in brain structure, even in light of individual differences, this will enable a new class of approaches for diagnosing disease and hopefully catching small changes at early stages in brain disease to address the problem early. I am excited about the possibility of leveraging information about the brain’s architecture to build and inspire the design of the next generation of deep learning architectures.

CP: Neuroengineering holds tremendous potential for developing new methods to help treat brain injury and disease. Traditionally, the options we’ve had for treating brain disorders are either medication or surgery. There’s a growing recognition that biomedical engineering solutions can play an important role here. We already have a rich history of using electrical stimulation in the brain, including deep brain stimulation for disorders like Parkinson’s disease, and cochlear implants to restore hearing for people who are deaf. Now we’re seeing emerging applications of closed-loop brain stimulation to epilepsy, depression, psychiatric disorders, and memory. We also have a rich history here at Emory. Mahlon DeLong, W.P. Timmie Professor of Neurology at Emory School of Medicine, who has been at Emory for decades, has laid down much of the scientific basis for how deep brain stimulation might be effective in Parkinson’s disease. Neurologist Helen Mayberg, who was here for quite a while before recently departing for University of Southern California, has been a pioneer in applying deep brain stimulation to other disorders like chronic depression. So I’m really excited by the team we have at the Emory Neuromodulation Technology Innovation Center (ENTICe), which brings together researchers from neurosurgery, neurology, and biomedical engineering, and also partners with folks at the GT Neural Engineering Center. We have an amazing team in place to develop the next generation of neuroengineering therapies for brain disorders.

If you wanted to convince a former colleague to move to Atlanta, what would be your top reasons?

CP: In terms of science, I love the community. As a member of the biomedical engineering department that spans Emory and GA Tech, I get to interact with wonderful colleagues at two very different institutions. Everybody here is extremely friendly and supportive. People genuinely want to interact and build collaborations and community and there’s not even a hint of competitiveness.

ED: I feel lucky to be part of a thriving community of neuroscientists, neuroengineers, and data scientists in Atlanta. Being part of both Georgia Tech and Emory provides exposure to a wide range of perspectives, plenty of opportunities for growth and collaboration, and what I perceive to be a lot of momentum around computational approaches to neuroscience.

If you could ask any person, dead or alive, to join your lab for a day, who would you ask?

ED: Richard Feynman. He was not only a brilliant scientist but also a dedicated teacher and mentor. It would be so cool to see him in action. He also created these diagrammatic representations of the behavior of subatomic particles called Feynman diagrams. Would be cool to see how he might think about visualizing and interpreting the types of problems that we are working on.

CP: One of the concepts we use in our lab a lot is called “dynamical systems theory”, which is a mathematical framework for understanding how complex systems behave. I think these approaches are key to understanding how the complex networks of neurons in the brain work. So I’d probably get a world-expert in dynamical systems, someone like Professor Steven Strogatz at Cornell, and force him to listen to me drone on and on about neurons until he caves and just tells us how to solve the brain.

A bonus question for CP: You’ve compared the behavior of neurons to the flocking of birds. How would you describe your research to someone who had no background in BME or neuroscience but doesgo bird-watching?

CP: So the activity of neurons and flocking behavior of birds has a little more in common than you might think. When you look at a flock of birds, you see these beautiful patterns pop out, and with complex, coordinated behaviors. Yet, there’s no one leader coordinating all of their activity and telling them what to do. You just have a bunch of individuals doing their own thing, but through very basic interactions, the group as a whole does something that’s tightly coordinated. This is what’s called emergent behavior, where putting relatively simple elements together can result in complex and coordinated activity. Neurons are very similar! You have a bunch of individual neurons that behave in certain ways, but the fact that they’re all wired up together and interacting with each other results in this coordinated, emergent system that is capable of doing really profound things. Another similarity here is that if I only showed you what one bird in a flock is doing, you’d have a hard time saying what the group as a whole is doing. But as you zoom out and start to observe more and more elements, the underlying patterns become clear. We think (and hope!) the same is true of neurons. My lab is monitors the activity of lots of neurons. Rather than focusing on what any one individual neuron is doing, we try to understand the coordinated activity of the network, and we try to really concretely understand how that network-level activity relates the brain’s control of behavior.

Courtney Shin