Understanding our brain

What is the HIBALL project about?

Katrin Amunts: HIBALL stands for Helmholtz International BigBrain Analytics and Learning Laboratory. It is an International Lab of the Helmholtz Association in collaboration between Forschungszentrum Jülich and McGill University in Montreal, Canada. As the name suggests, HIBALL is the continuation of our joint BigBrain project. We are now expanding this ultra-high-resolution model of the human brain, which we developed together more than ten years ago, to include information on the interconnection of cells and their molecular properties. With the enormous amount of data we are working with, we are using more and more tools from the field of artificial intelligence. However, the reverse is also true - how can we develop new AI tools around this brain model, based on the architecture of the brain?

Alan Evans: Some researchers look at the brain at the microcircuit level. That's very important. But I'm more interested in the overall organization of the brain and how it works as a whole. Large-scale regions "talk" to each other, characterizing the systems level of the human brain. In the HIBALL project, we're bringing together all these different spatial scales from molecules to individual neurons, to brain region, to understand how the brain works. And we're exploring how we move from one scale to the next. It's a fascinating problem.

How did you collect the data?

Katrin Amunts: We have been working on this project for more than ten years. The first step was to make 7,404 sections from a complete postmortem brain, stain the sections, and then reconstruct them. At the beginning, it was not at all clear how this could be done, because there were no methods yet to reconstruct a brain three-dimensionally with a resolution of 20 thousandths of a millimeter. It took several years to digitize all the slices alone. The reconstruction was a big step, because the original data set was a terabyte in size – an enormous amount at the time. And from that, we not only worked together to create a 3-D reconstruction of the brain, but also made it available to the scientific community.

And is it being used?

Katrin Amunts: Researchers around the world are using it. For example, a colleague in Marseille has used it to improve individual brain models of patients undergoing surgery for epilepsy. Such models are increasingly becoming an important tool for surgeons to better evaluate where and how much tissue to remove. If too much is removed around the epileptic focus, the patient may have a neurological deficit. If too little is removed, the epileptic seizures may persist. With BigBrain, surgeons can integrate information about the distribution of neurons into a mathematical brain model that is also enriched with their patients' data - creating a personalized model.

And now HIBALL comes into play?

Alan Evans: Right. HIBALL is the next chapter. BigBrain shows the distribution and architecture of the cells in the brain; with HIBALL we add more data. So now we are also looking at the chemistry of the brain, which is very important for how signals are passed from one cell to another. We are also analyzing connectivity, i.e., the connective structure of the brain. So, we are combining very many different types of information. It's a bit like Google Earth, where you can overlay country borders, roads or vegetation zones on top of the map of the earth, for example. The inferences you draw from using each of those overlays are quite different, even though you used the same planet as a template. And then we use tools like artificial intelligence to examine all that data and turn it into knowledge

What kind of AI tools are you using?

Katrin Amunts: For example, we use Deep Learning to identify brain regions in the BigBrain. To do this, we train an AI algorithm to recognize the individual cells of the brain. We are interested in how the cells are distributed. This is characteristic of each brain region, and we then explore what that has to do with the function of one of those brain regions. So Deep Learning helps us ask those kinds of questions and characterize the BigBrain in detail. On the other hand, however, via the BigBrain, we also learn how cells and their networks are organized in our human brain. These natural networks have many strengths. For example, we humans are very good at recognizing faces, even when they are partially obscured or altered. Artificial neural networks sometimes have great difficulty in correctly assigning faces under such conditions. We now want to know what specific properties of our natural brain networks make them so powerful in this area. On the other hand, our brain is not particularly fast at multiplying two very large numbers together - a computer can do that many times faster. Why is that? We want to learn the rules by which the brain works, but also how to build more powerful artificial neural networks. So HIBALL is really the next step after BigBrain.

That sounds like a task for a multidisciplinary team.

Alan Evans: That's right. HIBALL is also Big Data, for example. It's a project that brings together people from many different disciplines: Computer scientists, mathematicians, neuroanatomists, neurologists, cognitive neuroscientists, AI specialists. They are all part of the project, and we need them all. It is no longer a descriptive science that only observes. Rather, it is a quantitative science. That means we need the whole machinery of data science to understand connectivity in the brain. And to understand how it changes during development. And how it breaks down in disease. The great joy but also the challenge for Katrin and myself is to keep this "menagerie" together and make sure they are all looking in the same direction as a team. Fortunately, we work with colleagues who are both scientifically excellent and great people. We have a lot of fun together. And that makes all the difference in an international project where you have to work together while separated from each other.

And what about you two? What is your scientific background?

Alan Evans: I consider myself kind of an outcast physicist. [laughs] I trained in mathematics and physics a long time ago, and I did a PhD in protein crystallography to study the structure and function of proteins. And then somehow, I came across the structure and function of the human brain – a difference of around 10 orders of magnitude.

Katrin Amunts: I have a different background. I studied medicine and biophysics. Later, I focused more on the brain and used image analysis to capture the relationship between brain structure and brain function.

The brain has become your life's work. What fascinates you most about it?

Alan Evans: From the time I was ten years old, I was fascinated by human brains and memory. I remember reading about it in Scientific American. I never thought I would be working at the Montreal Neurological Institute alongside a great scientist like Brenda Milner, who is one of the leading minds in our field. It used to be that descriptive characterization of the brain by region was the scientific standard. Then we moved to a model of the brain that looked at the connections between those regions. That changed the way we think about the brain. I now want to go one step further. I want to know what the ideal spatial scale is to understand the organization of the human brain.

Katrin Amunts: At the beginning of my career, I particularly wanted to understand how the brain contributes to our movements and language. What does the organization in the human brain look like that enables such complex functions? Why can we speak and others cannot? I was fascinated by these questions from the beginning. I mapped areas in the language region of the brain and realized that the complexity of the brain is related to its organization at multiple levels. And I want to understand how to make predictions from one level to another. Let's look at a particular cellular architecture: what does it mean for function? What does it mean for function in a huge network? What does it mean for a patient if that region is damaged by a stroke, for example? I'm very interested in understanding how to build bridges between the different spatial scales. And it is indeed the case that the more you learn, the more you realize how much is still missing. And that's also what makes brain research so exciting.