To carry out their activities, Research Teams of the Frédéric Joliot Institute for Life Sciences have developed high-profile technological platforms in many areas : biomedical imaging, structural biology, metabolomics, High-Throughput screening, level 3 microbiological safety laboratory...
All the news of the Institute of life sciences Frédéric Joliot
Scientific result | Brain | Cognition | MRI | Artificial intelligence
In this study, a NeuroSpin team succeeded in decoding a large set of human mental processes by training neural networks to predict cognitive labels from a cognitive atlas and by leveraging the data set of NeuroVault, a web-based database for collecting and sharing statistical maps of the human brain. A success that now makes it possible to infer a mental process from observed brain activity.
To better understand the links between the cognitive (mental) processes responsible for the treatment of all the information we exchange with our environment and the global functional organization of the brain, scientists analyze "brain maps" established by functional magnetic resonance imaging (fMRI). The latter has opened up the possibility of studying, through statistical analyses, how and where our brain activity is modulated in the many mental processes of our daily lives. It is now common to introduce fMRI data into machine-learning models, in order to encode brain activity in response to a behavior, and then to be able to decode this activity, i.e. to draw conclusions about the function of a brain structure.
In this work, the researchers went much further by expanding the scope of the usual image-based studies : by using all statistical maps of fMRI imaging from the largest available data repository, NeuroVault, they trained machine learning models capable of decoding the cognitive concepts probed, including in novel studies that may involve previously unexplored experimental conditions. To do this, they labeled NeuroVault images with concepts from the Cognitive Atlas, an ontology of cognition, by tapping into the metadata associated with the atlas. They trained neural networks to predict these cognitive labels on tens of thousands of brain images. In this way, they were able to decode more than 50 classes of mental processes, while overcoming the heterogeneity, imbalance, and noise of the training data, without any prior knowledge of the experimental setting or the relevant concepts. These results demonstrate that image-based meta-analyses can be undertaken on a large scale and with minimal manual data processing. They allow broad reverse inferences to be made, i.e., to conclude about mental processes from observed brain activity.
Contact : Bertrand Thirion (email@example.com) Website : https://team.inria.fr/mind/
Romuald Menuet, Raphael Meudec, Jérôme Dockès, Gaël Varoquaux, Bertrand Thirion. Comprehensive decoding mental processes from Web repositories of functional brain images. Scientific Reports, (2022) 12:7050 https://dx.doi.org/10.1038/s41598-022-10710-1
CEA is a French government-funded technological research organisation in four main areas: low-carbon energies, defense and security, information technologies and health technologies. A prominent player in the European Research Area, it is involved in setting up collaborative projects with many partners around the world.