The associated team led by Bertrand Thirion (Inria project-team Parietal). This project brings together computer scientists and neuroscientists with varied and complementary skills. Between coding sprints and tool development, the Meta&Co team focuses on the analysis of very large-scale brain images.
Known as MetaMRI between 2015 and 2017, the associated team became Meta&Co, Machine learning for meta-analysis of functional neuroimaging data, in 2017 and ended in 2020. This associated team is the result of a collaboration between the Parietal project-team of the Inria Saclay – Île-de-France centre, specialised in the statistical modelling of brain functions using neuroimaging data, and a laboratory at Stanford University, the Poldrack Lab, based in the university’s psychology department.
The two partners have been working together on very large-scale brain imaging analysis issues. Russel A. Poldrack’s team at Stanford has developed public resources for collecting brain image data, documenting brain function and allowing it to be studied. This has enabled Parietal to develop analysis methods to exploit the richness of this data, using large-scale statistical learning.
Bertrand Thirion, head of the Meta&Co team, tells us more about the results and prospects of this associated team.
What was the point of proposing this associated team?
First of all, this associated team gave a formal framework to the collaboration and an associated budget allowing the implementation of various actions and meetings, at Saclay as well as at Stanford, necessary for our research.
Thus, various developer meetings, coding sprints and other events have taken place. These are crucial meetings to bring new ideas to the project. This associated team has created opportunities to work together. It has allowed us to share ideas, tools, data and analysis techniques.
What has been the impact?
The framework of the associated team has been very useful on several levels.
First, the collaboration had an impact at the conceptual level. As computer scientists, we do not have all the keys to address all the issues related to brain knowledge. Therefore, working with cognitive scientists and psychologists has been very interesting. It allowed us to ask questions that are relevant to current knowledge and data related to the brain.
Then there is the access to and understanding of the data, which has led to several publications. Some of these articles* have appeared in prestigious journals such as Elife and PLOS Computational Biology . These scientific journals specialising in biology, computational biology or medicine, for example, have enabled us to have a greater impact than in the more traditional publications that we produce in our discipline.
Finally, the associated team had a real impact in terms of software. Alongside the Poldrack Lab team, we have contributed to the implementation of open source software tools for the community. On this occasion, we organised coding sprints to develop our tools. These meetings were a source of exchange and collective emulation.
Dialogue is the key to understanding how best to build on what exists. Bertrand Thirion
These benefits are closely linked to the diversity of skills of the actors involved within Meta&Co. It is this complementarity that has enabled us to achieve significant results.
What were the results?
In addition to the various publications mentioned above, we have developed digital tools related to the analysis of brain images.
Within the framework of this collaboration, we have created a web interface called “neuroquery.org “. This tool offers users the possibility of obtaining, following a query relating to cognitive sciences such as attention during reading , a representation of the brain that reflects what the existing scientific literature on this subject states.
Web interface developed by Meta&Co, https://neuroquery.org/
We also developed a “functional atlas” of the brain , in order to identify the brain areas necessary to perform certain everyday cognitive tasks (movement control, sensory perception, learning, decision making, language, etc.); after systematic analyses of the available data associated with cognitive labels. We were thus able to identify and share significant associations between certain parts of the brain and cognitive terms. We are the first to have published systematic atlases of these associations between cognitive science concepts and brain images.
Finally, on a personal level, this collaboration allowed me to launch my IA Karaib (Knowledge And RepresentAtion Integration on the Brain) chair , the main objective of which is to develop an automatic classification model to link data from functional brain imaging and descriptions of behaviour or cognitive diseases.
What are your prospects now?
Although the associated team is officially closed, we will obviously remain in contact.
In fact, the Karaib AI Chair will continue this work and develop some of the aspects addressed by the associated team. As a continuation of our research, we now want to gather and analyse the various contributions, whether they are based on text or images. In this way, we wish to merge neuroquery.org with the analyses conducted in parallel on image databases.
This chair will have an undeniable impact on the post-collaboration. It will also be able to cover some of our expenses, such as visits by our students to Stanford.
What is your assessment of this experience?
Firstly, it is important to stress that the administrative burden of launching an associated team is reasonable. Secondly, I think it is a sign of good health to have an associated team, as it shows an openness towards the international community and sometimes towards new disciplines. It also represents a real opportunity to increase the impact of our work.
In the framework of an associated team, we seek complementarity through capacities that differ from our own. This can broaden the scope of the project team involved by giving it access to new skills. Finally, it allows us to have a greater impact by addressing other communities, needs and ways of working.
* NeuroQuery, comprehensive meta-analysis of human brain mapping , Jérôme Dockès, Russell Poldrack, Romain Primet, Hande Gözükan, Tal Yarkoni, Fabian Suchanek, Bertrand Thirion, Gaël Varoquaux. Elife , March 2020.
https://hal.archives-ouvertes.fr/hal-02485642v2
Atlases of cognition with large-scale brain mapping , Gaël Varoquaux, Yannick Schwartz, Russell Poldrack, Baptiste Gauthier, Danilo Bzdok, Jean-Baptiste Poline, Bertrand Thirion. PLOS Computational Biology , November 2018.
https://hal.inria.fr/hal-01908189
Identity card of the associated team
- Name: Meta&Co(Machine learning for meta-analysis of functional neuroimaging data)
- Inria team: Parietal
- Partners: The Poldrack Lab, a Stanford University laboratory based in the Department of Psychology
- Project: Contribute to the development of tools for the advancement of very large-scale brain imaging analysis
- Keywords: brain, cognitive science, brain imaging, analysis
——————————————-
To find out about the other Associate Teams of Inria@SIliconValley
