Micrometre by Micrometre: Decoding the Human Brain with AI

The starting signal has been given: the German-Canadian Helmholtz International BigBrain Analytics and Learning Laboratory (HIBALL) takes up its work. The goal: a three-dimensional brain atlas at a cellular resolution level. The method: the close integration of artificial intelligence, supercomputing and neuroscience. Involved are: more than 40 scientists.

In 2003, when Jülich neuroscientist Katrin Amunts and her Canadian colleague Alan Evans began scanning 7,404 histological sections of a human brain, it was completely unclear whether it would ever be possible to reconstruct this brain on the computer in three dimensions. At that time, there were no technical possibilities to cope with the huge amount of data. The researchers did not get discouraged, however, and tackled it – BigBrain was born. Today, scientists use the 20-micrometer model as a reference brain – the technical foundations have long been laid.

The HIBALL research cooperation is now going one step further and aims to develop a brain model with an accuracy of one micrometre, that is, one thousandth of a millimetre. Several petabytes of neuroscientific data have to be processed – it has still not been resolved exactly how. However, Katrin Amunts and her colleagues are again relying on technical progress. At HIBALL, therefore, disruptive procedures from artificial intelligence (AI) will for the first time take on a major role.

Prof. Katrin Amunts is director of the Institute of Neuroscience and Medicine (INM- 1) at Forschungszentrum Jülich as well as the C. and O. Vogt Institute of Brain Research at Heinrich-Heine University Düsseldorf. She is the Scientific Research Director of the Human Brain Project. Forschungszentrum Jülich / Sascha Kreklau

An interview with Prof. Katrin Amunts about the challenges and significance of HIBALL for the international research community.

In the USA and Canada, “highball” was also a term for a signal for trains rushing through a station at full speed. Is it a metaphor for the German-Canadian brain research cooperation, which is now officially gathering momentum?

Amunts: Yes, we did indeed like this picture, and it played a role in the naming of the project. In a way, it stands for “let’s go ahead – at full speed”.

You’ve been working successfully with your Canadian colleagues – with Prof. Alan Evans in particular – since the 1990s. What makes the International Lab HIBALL so special for you?

The research cooperation takes collaboration to a new level, one that goes beyond the Jülich-Montreal bilateral axis. We now also have partners from AI and supercomputing on board to further develop the BigBrain brain model. Eventually, a reference brain is to be created which will form the basis for many other questions – for example in medicine.

The resolution in the BigBrain model is 20 micrometres. At HIBALL, you now want to create a brain model with an accuracy of one thousandth of a millimetre. Why?

At 20 micrometres, you might see most of the cells. However, there are also cells that are only ten micrometres in size, and these have previously remained “blurred”. Therefore, within each tissue section, we want to go to the level of one micrometre. The goal will be a spatial resolution at 1 x 1 x 1 micrometre. Only then can the cells with their different forms and extensions be seen and it can be understood how they are arranged in the brain. On this basis, brain functions and cognitive performance can be associated.

Ultra-high -resolution 3D maps of cytoarchitectural areas in the Big Brain model. Users can navigate in the Big Brain, zoom in to get more details, and see how the brain is parcellated.
Amunts, Schiffer, Dickscheid et al.

This extremely high resolution goes hand in hand with gigantic amounts of data. How do you intend to cope with these masses?

It’s a great challenge. At HIBALL, we talk about several petabytes. This can only be achieved by using completely new analytical methods – for example those that work with deep neural networks or machine learning – modern memory and communication technologies, and powerful computers. In addition, we want to analyse networks of nerve cells in the human brain and develop network models in order to deduce how the function of artificial neural networks can be improved.

So it’s a win-win situation for neuroscientists and AI experts?

Yes. On the one hand, AI and computing researchers benefit from BigBrain as well as the related knowledge and models of neural networks. On the other hand, we neuroscientists benefit from AI and computing. One thing is clear: we can no longer calculate and analyse the huge amount of data without the modern methods.

Is AI the key to understanding the entire brain one day, then?

I is certainly one of the keys. At HIBALL, we use machine and deep learning methods, which both count among AI. Without them, we could not tackle such an ambitious project. That’s why we are pleased to have CIFAR as a partner, the research organization that is leading the Pan-Canadian Artificial Intelligence strategy. With MILA in Montreal, we got one of the world’s leading centres in the field of deep learning on board. Its scientific director, Yoshua Bengio, is one of the pioneers in this field.

Do you need deep learning methods in your everyday life as a neuroscientist?

Today, the amount of data in one single image and the complexity of the information are often so enormous that no scientist can analyse them “manually”. Automated and robust methods are needed to process and analyse these large image quantities. In many areas, deep-learning methods are increasingly supplementing classical image analysis methods. We neuroscientists see that artificial neural networks are extremely helpful in recognising certain patterns or in calculating three-dimensional brain models. Due to the artificial neural networks, the tools to solve such methodologically difficult questions have changed considerably in recent years.

Welche Rolle spielt in diesem Zusammenhang die Supercomputing-Expertise aus Jülich?

The combination of supercomputing and the deep learning methods at both Jülich and Montreal is particularly promising. With the volume of our brain data, we quickly reach capacity limits – so you need supercomputers, a lot of storage space and experts on site. This is what Jülich offers.

Plus: at HIBALL, we want to exchange large amounts of data between Canada and Germany – but there are a few thousand kilometres between them. For this reason, we want to create a common platform that is suitable for this exchange. There’s a lot of technological know-how in it, which the colleagues from the JSC bring with them. Together with their Canadian colleagues, they are laying the groundwork for cooperation and creating a platform that will connect the world: in the future, scientists from all countries are to use the HIBALL platform to calculate and research on the same data together, to exchange, download or upload it.

Today, the BigBrain model and the 3D maps are a prime example of shared big data. They can now be clicked on, rotated, zoomed in and marvelled at by anyone on the Internet. Is something similar to emerge from HIBALL in the end?

In principle, yes. When it comes to the 20-micrometre BigBrain model, we’re talking about one terabyte of data – even that’s not something that can “just quickly be downloaded”. That’s why we have developed web-based tools in which only the data you are looking at has to be transferred. This is similar to what has already proven successful in other areas. If we go to the 1-micrometre level, it’s going to be even more challenging. So we need to develop methods that allow data to be processed without having to transport it.

Left: Cortical layer segmentation in the BigBrain (Wagstyl et al.), visualized in the HBP atlas viewer. Right: A neural network mapping the brain. The different layers of the network capture different tissue properties.
Amunts, Schiffer, Kiwitz, Dickscheid et al.

Neuroscientists, computer scientists, mathematicians – how do you even manage to speak the same “language” with your colleagues from AI and supercomputing?

This has never been easy, really, but interdisciplinary work is the essential basis for success. I think that this successful work across disciplines is indeed one of the greatest achievements of recent years. We also profit from this at HIBALL. We have common visions, there is a lively exchange – not only on paper. We live this cooperation. I myself might grab my mobile phone and just call!

At HIBALL, however, we also want to ensure that the next two generations of researchers will be qualified at the border between neuroscience and computing. We therefore support the transatlantic exchange of researchers at all career levels at HIBALL.

What is the significance that HIBALL has for Forschungszentrum Jülich?

The lab shows that innovative research is being conducted at Jülich across national borders in a particularly rapidly developing field of research. The Helmholtz Association funds only a few international labs – only three in 2018. The cooperation is also an important contribution to Forschungszentrum Jülich’s strategy, in which information and data sciences play a central role. With HIBALL and artificial intelligence, neuroscience has now set another strong priority.

What role does HIBALL play with regard to the European flagship initiative, the Human Brain Project, of which you are the scientific director?

HIBALL complements the development in the Human Brain Project, but also the Canadian initiative “Healthy Brains for Healthy Lives”. With these partnerships, HIBALL can have a far greater impact than if it acted alone, and the partner institutions also benefit from the strengthening at the interface between spatial high-resolution brain models, computing and AI. It’s of great advantage if one can bring about that many small and large cogs interact and develop a great dynamic. HIBALL will benefit the two major initiatives. We’re currently seeing how neuroscience is benefiting from the revolution in information technologies. HIBALL is to provide important impulses which will have an impact on neuroscience, medicine, and also on technology development and society.

A personal final question: you were born in 1962. Back then, during your studies, would you have thought that one day you would be able to “zoom in” on the world of the brain?

There has indeed been an incredible change since my student days, when I was already interested in image analysis to better understand the architecture of nerve tissue and its functions. At that time, we had an image analysis system from Leitz at the institute that made it possible to quantify and statistically describe the architecture of nerve cells. Back then, this seemed to me the right way to understand the organising principles of the brain, and this research shaped my further path.

When we started making histological sections for BigBrain in 2003, we didn’t have a concrete plan as to how exactly we could digitise the 7,404 images and reconstruct them in three dimensions. There was an infinite number of single problems that had not yet been solved. So we simply trusted that we would be able to create the technical conditions one day. And that was right! The cooperation with our partners in Montreal, with Alan Evans and his team, has led to a breakthrough in this.


The official starting signal has been given on 26 June 2020: with the Helmholtz International BigBrain Analytics and Learning Laboratory (HIBALL), a German-Canadian research consortium with partners from the Helmholtz Association, from McGill University, MILA and CIFAR has been inaugurated in Montreal. Scientists from all over the world took part in the online conference – among the speakers at the opening are the German Ambassador to Canada, Sabine Sparwasser; Prof. Otmar Wiestler, President of the Helmholtz Association; Prof. Wolfgang Marquardt, Chairman of the Board of Directors of Forschungszentrum Jülich and Vice-President of the Helmholtz Association; and Pawel Swieboda, General Director of the Human Brain Project.

HIBALL is funded with a total budget of € 6 million by the Helmholtz Association and the Canadian partners for a period of five years. Montreal – with MILA as one of the world’s leading centres for deep learning – is an important location for the project’s objectives. A total of around 40 scientists from Jülich and Montreal are working together on the highly detailed 3D model.

The interview was held by Katja Lüers

Further information:

Last Modified: 07.09.2023