You are here : Home > Research entities > Research themes

Research themes

Parietal


Bertrand Thirion is in charge of this project.
Published on 28 November 2017

The Parietal team focuses on mathematical methods for statistical inference based on neuroimaging data, with a particular reliance on machine learning techniques and applications of human functional imaging. This general theme splits into the following research axes:

  • Mathematical methods for multi-modal brain atlases are a means to expose the information available from images as a template informed by different observation modalities, and to account for the standard variability around this template. The Parietal team focuses on improving the underlying estimation procedures, and developing new markers, such as those derived from functional connectivity models.
  • Statistical analysis for high-dimensional data addresses the variability of brain structure and function in a statistical perspective. Parietal aims at contributing statistical models to detect populations effects with enough sensitivity by exploring recent (e.g. random forests) or novel (multi-task learning) statistical methodologies.
  • Modeling brain function through neuroimaging addresses another key question, namely the modeling and understanding of brain function based on functional imaging measurements. Parietal’s contribution to encoding models will address the particular case of vision, a system which can be observed accurately with MRI, given its size and the resolution achieved with current MRI scanners
  • One of the major challenges in encoding, i.e. the modeling of brain function through neuroimaging is the neuro-vascular coupling which is inherent to the BOLD signal. This has been addressed in the so-called Joint Detection Estimation framework, which is continuously improved, both on the computational and modeling size, e.g. with the inclusion of spatial models and new imaging modalities.
  • Parallel MRI acquisitions techniques with multiple coils have emerged as fast powerful imaging methods, in which the image data is reconstructed based on partial acquisition of the 2D or 3D Fourier transform of the image or volume, respectively. The next step being to unify SENSE and compressive sensing approaches to gain additional acceleration, our aim is to achieve high-performance acquisitions to optimize the use of scanning time without loss of quality.

Parietal is also strongly involved in open-source software development in scientific Python (machine learning) and for neuroimaging applications. Specifically, Parietal is the main contributor of the Scikit-Learn, mayavi, Nilearn and PyHRF free software.

See https://team.inria.fr/parietal/