Cloud computing for research in neuroimaging
The Parietal team is taking part in the Azure Brain project, in collaboration with Microsoft and the joint Inria-Microsoft Research research laboratory, to analyse large masses of neuroimaging data through cloud computing. Bertrand Thirion, the team's leader, explains the potential benefits of this innovative project.
It is essential to understand the variability among the brains of a population of individuals, in order to better appreciate the normal differences, but above all to properly diagnose certain illnesses affecting the central nervous system. Much of the variability is already present at birth, and it is probably linked to our genetic code. It is therefore particularly important to know how to compare the genes of an individual and the characteristics of his/her brain, such as those observed on a functional or anatomical MRI.
In recent years, a lot of work has been done firstly to compare genetic information with behavioural data or the presence of a psychiatric or neurodegenerative illness, and secondly to compare brain images with this kind of information. It would now appear to be important to link genetic data directly with brain images, because these images represent an intermediary characteristic which could make it easier to link the differences observed between genetics and behaviour.
However, the comparison of neuroimaging data and genetic data poses considerable problems for statistical analysis. We do not know which genes are linked to a given signal in a given region of the brain. As part of this approach, it may be useful to systematically test the links that exist between genetic variables and neuroimaging variables.
The use of cloud computing for these calculations is a precious tool, as it allows large-scale experimentation on giant databases.
But the statistical problem is then limited by the performance levels of the computers, as we have a million variables in each field... It is necessary to test the meaning of each possible link, taking care to ensure that no false detections are made, which demands considerable computing resources. On the other hand, the problem can easily be parallelised, as it is a matter of repeating the same elementary sequence of calculations with different data.
The use of cloud computing for these calculations is a precious tool, as it allows large-scale experimentation on giant databases. By using cloud computing , we hope to facilitate the discovery of interactions between certain genes and certain differences in the functioning or shape of the brain.