Tools for getting more out of brain imaging
Brain imaging has made a great deal of progress in recent years, providing more and more data and with an increasingly high level of quality. Developing reliable, high-performance tools to improve the use of these images is now vital. Inria has an important role to play in this respect. The Parietal team is presenting four articles on this subject at MICCAI 2012, as well as contributing to four specialist workshops and running a demonstration at the Inria stand.
Bertrand Thirion, Parietal team leader, Inria Saclay–Île-de-France
What have been the big advances in brain imaging in recent years?
Brain imaging has made a lot of progress thanks to improved equipment performance and the development of increasingly sophisticated image analysis techniques. The advent of high-field MRI makes it possible to obtain excellent images of brain activation. Today, the resolution of such images is 1 mm, as against 3 mm just 5 years ago. The data are therefore of higher quality and enable more detailed study of the functional connections between the different regions of the brain.
This study is facilitated by the development of new techniques for analysing these images, which are producing increasingly precise models of cortical connections. These techniques are based upon fairly complex calculation methods which still require improvement.
A third important aspect of this research field is currently being developed with the emergence of databases containing immense quantities of data about the brain, which are only growing as the years go by. I think that a major challenge for the future is to create the tools necessary to manage these imaging databases.
What kinds of tools are necessary to make use of these huge databases?
We need to develop computerised tools that allow us to reuse these data, such as for meta-analysis. This is useful, for example, when we wish to compare data obtained via different protocols or to combine data from similar experiments in order to confirm a result or establish more precise and more objective hypotheses. The main difficulty involved in producing these kinds of tools is knowing how to manage the information associated with the images, particularly the protocol used to obtain them. Once the data have been organised, learning technologies come into their own by automatically identifying characteristics common to a large number of data items. At the Inria stand at MICCAI, our team will present a demonstration of a software program that offers the tools necessary for such an approach: scikit-learn.
Another challenge for computer scientists is to be able to use images from a different modality (anatomical, functional, etc.) to offer fuller and more precise information about a part of the brain. At MICCAI, we are presenting a project concerning ways of pooling two sources of information: one supplied via functional imaging, concerning dynamic formation of activation networks, and the other provided by the diffusion MRI, concerning the fibres connecting the various regions of the brain. The goal is to combine the two types of images to establish a link between the main fascicles of brain fibers and the activation networks and thus build a better model. It is also possible to use one method to guide interpretation of another.
Can we envisage medical applications of these advances?
In the medium term, this work will be used to study patient cohorts - for example, to identify risk markers for neurodegenerative diseases with a view to diagnosis and prognosis.