Prof. Katrin Amunts was the head of the EU Flagship Human Brain Project. The project laid the foundation for a new brain atlas and personalized brain models, which are now available via EBRAINS. Such brain models can be used to plan tailored medical treatments. We spoke to her in this interview.
Researchers have been aware of individual differences for more than a hundred years and have also described them – variations in the thickness of the cerebral cortex, its folding, or the arrangement of cells. It was also suspected that these differences could be relevant for brain function or disease. However, it was not until the late 1980s that functional imaging made it possible to observe the brain “at work”, so to speak. This raised more fundamental questions, such as: which region of the brain is associated with a specific function? In other words, which area is activated when I move my finger, and does this happen in the same place for all people?
All brains differ slightly in their anatomy, and at the time, such natural variability was often seen as disruptive noise. In medicine, however, the focus is on the individual patient, and so the concept of personalized medicine was established. This is an important development to which we contribute in neuroscience – for example, by developing personalized brain models.
There are different types of brain models – anatomical, biophysical, data-driven models, and others. Some brain models describe the processes in the brain mathematically, such as “The Virtual Brain”, which was developed by Viktor Jirsa and his team from Marseille in the Human Brain Project and allows certain brain processes to be simulated. It enables parameters to be changed; the virtual brain can therefore be personalized. This includes findings such as the patient’s MRI scans, their electrical brain activity, and information about their connectivity structure. These are supplemented by data from brain atlases, which provide fundamental parameters such as cell densities. The end result is a personalized model based on specific patient data. Achieving this was a major advance.
In France, a clinical study on epilepsy is currently being completed. The patients suffer from seizures that originate from specific, individual brain regions. These regions can be surgically removed. It is important to determine their location as accurately as possible. Using data on each patient’s electrical brain activity and structure, our colleagues developed a personalized model for each patient. The model makes it possible to calculate how much tissue needs to be removed for each individual and from which sites in order to stop the seizures without causing damage. Our colleagues are working on making such models more informative, more individualized, and more precise, as well as applying them to other areas.
Our brain atlas describes the probability of finding a specific area at a given location in the brain. It takes anatomical differences into account, in other words, the variability of the brain. We began developing this atlas here at the institute and at the University of Düsseldorf in the early 1990s. To do so, extremely thin slices of donor brains were prepared and stained to make the cell bodies visible under the microscope. We then digitized the slices, reconstructed each brain in 3D on the computer, and mapped individual areas of the brain using statistical methods. The individual maps were then superimposed to create probability maps – the basis for the Julich Brain Atlas, which for the first time captures variability in such a spatially precise way. It is something that only exists at this resolution, level of detail, and scale at Jülich. At the same time, it forms the core of the EBRAINS digital research infrastructure – a product of the Human Brain Project.
We have always placed great value on ensuring that tools and data are easily accessible and can be used in accordance with FAIR data principles.
This is possible via EBRAINS. Scientists around the world can use the platform to access data and digital tools for neuroscience. They can use the atlas to solve their own problems – whether these relate to basic research, cognitive research, or medical issues. We have always placed great value on ensuring that tools and data are easily accessible and can be used in accordance with FAIR data principles. And this is increasingly happening now.
We are currently developing “brainfm”, a foundation model of the human brain at the cellular level. “brainfm” will be significantly more powerful than conventional AI models. In future, it will be used in basic research as well as in personalized medicine. The model represents approximately 86 million neurons and a similar number of glial cells, which support the neurons. It also has a resolution on the scale of 1 micrometre – one thousandth of a millimetre. The model thus comprises two to three petabytes of data – an enormous amount. To compute such a model, JUPITER is required. We are very pleased that “brainfm” is one of the first 18 software codes to run on JUPITER. This helps us to make rapid progress in our research.
This text is taken from the 2/25 issue of effzett. Text: Artur Denning
