Meta&Co (Since 2015)
Machine learning for meta-analysis of functional neuroimaging data*
*formerly MetaMRI (from 2015 to 2017): Machine learning for meta-analysis of functional neuroimaging data
- Bertrand Thirion, PARIETAL project team, Inria Saclay Ile- de-France research center
- Russ Poldrack, Stanford University, Psychology department
Meta&Co (since 2018) is an associate team that is the follow-up of MetaMRI (since 2015).
Cognitive science and psychiatry describe mental operations: cognition, emotion, perception and their dysfunction. Cognitive neuroimaging bridge these mental concepts to their implementation in the brain, neural firing and wiring, by relying on functional brain imaging. Yet aggregating results from experiments probing brain activity into a consistent description faces the roadblock that cognitive concepts and brain pathologies are ill-defined. Separation between them is often blurry. In addition, these concepts and subdivisions may not correspond to actual brain structures or systems. To tackle this challenge, we propose to adapt data-mining techniques used to learn relationships in computational linguistics. Natural language processing uses distributional semantics to build semantic relationships and ontologies. New models are needed to learn relationships from heterogeneous signals: functional magnetic resonance images (fMRI), on the one hand, combined with related psychology and neuroimaging annotations or publications, on the other hand. Such a joint effort will rely on large publicly-available fMRI databases shared by Poldrack Lab, as well as literature mining.
- The Associate team actively contributed to setting up the NeuroVault infrastructure www.neurovault.org and showing its relevance for cognitive neuroscience. NeuroVault is a web resource where cognitive neuroscientists can upload brain maps that represent the results of their experiments, together with annotations that make it possible to compare these data across studies. This resulted in two publications.
- A major contribution to the field has been the set up of common conventions for data organization. The Brain Imaging Data Structure (BIDS) initiative is a landmark project, launched by Poldrack Lab, but also contributed by Parietal. More recently, a thread of developments based on these BIDS conventions has attracted lots of contributions, such as the BIDS-App and Openneuro suite, with contributions from Parietal.
- B. Thirion and R. Poldrack have taken part to the formulation of recommendations for the neuroimaging community on the best way to run and report analysis pipelines (COBIDAS). This action is central to the purpose of MetaMRI: foster data exchange and large-scale data analysis across the community.
- An important consequence of the joint team has been the diffusion of the resources produced by the Individual Brain Charting (IBC) project, a project driven by Parietal to provide a quasi exhaustive mapping of cognitive functions in a limited set of participants: Parietal has shared a first batch of brain maps to the NeuroVault repository https://neurovault.org/collections/2138, and is currently uploading the raw data
to OpenfMRI; both resources are managed by Poldrack Lab.
- A work-in-progress action is the publication of the first large-scale analysis of multiple datasets, using the shared concepts across studies and a sophisticated two-layer classifier. This was the result of a French ANR project, but it has been further discussed and improved along the MetaMRI project and is submitted for publication. This work establishes clearly the impact of forward and reverse inference in cognitive brain mapping.
- Another ongoing action is the Neuroimaging Power project, that consists in measuring effect size for various cognitive effects (typically those described in the large-scale analysis above), based on the data available in the NeuroVault and OpenfMRI repositories. This aims at providing a reference for power analysis of neuroimaging experiments and improve statistical inference. This is made concrete through the Marie Curie Project of Joke Durnez between Poldrack Lab and is a consequence of MetaMRI.
- MetaMRI has also lead to a PhD project, called MetaCog, in which Parietal researchers leverage text mining techniques for the definition of relevant psychological concepts that are associated with brain activation maps. This project is done in collaboration with Poldrack Lab (the PhD candidate, J. Dockes, is currently spending two months in Stanford), as Poldrack Lab has done seminal work in this direction and is actively pursuing it. This is a very promising common direction of research.
Publications and Awards:
- 9 publications
- Software development:
The coding sprints held during the project have lead to major improvements to the software produced by the partners.
• Nilearn has made great progress, in particular thanks to the international coding sprints held thanks to MetaMRI, which has made it possible to add lots of features (improved visualization, continuous integration of the documentation, cleaning of he API and the underlying code), that has fostered the rapid adoption of Nilearn by the community (95 citations for the seminal paper).
• The OpenNeuro coding sprint has contributed to improving the neuroimaging data processing suite designed by Poldrack Lab, that is a major contribution to the set up of public and powerful infrastructure for data processing. Parietal has taken part to the coding sprint and given feedback on the current implementation. Parietal has also developed Python code that is being integrated into this framework.
• NeuroVault is a public repository of brain maps with a relatively simple API to upload brain images with annotations. This is lead by Poldrack Lab, but Parietal has contributed to the effort.
Krzysztof J Gorgolewski, Fidel Alfaro-Almagro, Tibor Auer, Pierre Bellec, Mihai Capotă, M Mallar Chakravarty, Nathan W Churchill, Alexander Li Cohen, R Cameron Craddock, Gabriel A Devenyi, et al. Bids apps: Improving ease of use, accessibility, and reproducibility of neuroimaging data analysis methods. PLoS computational biology, 13(3):e1005209, 2017.
Meta&Co (starting in 2018 for 3 years) is the follow-up of the associate team MetaMRI (2015-2017).