Learning to map text to locations in the brain
This project is supported by Digiteo. Jérôme Dockès is supervised by Gaël Varoquaux, Bertrand Thirion and Fabian Suchanek.
Neuroimaging studies often have a limited scope or lack statistical power. Meta-analysis, pooling many studies together, allows us to discover where there is a consensus in the literature. Through large-scale statistical analysis of thousands of studies, we can extract reliable knowledge from noisy data.
We build predictive models that map textual descriptions of experiments, mental processes or diseases to anatomical regions in the brain. A trained model can be used through a web interface to compute brain maps of the regions associated with a text query. A Python library enables training and querying text-to-brain mapping models.
To train these models, we collect and share the largest existing dataset of neuroimaging studies and brain activation coordinates. See Text to brain: predicting the spatial distribution of neuroimaging observations from text reports, or this presentation for details.
Decoding brain images from many different studies
For this project, Romuald Menuet is supervised by Bertrand Thirion and Gaël Varoquaux.
In this project we collect dozens of thousands of fMRI statistical maps of the brain, produced by hundreds of different neuroimaging studies from the Neurovault platform.
We then train machine-learning models to predict the mental conditions associated with an image of brain activity. Leveraging the huge amount of data available on Neurovault enables to distinguish dozens of different mental conditions.
