Computerized analysis of brain anatomy and functions has an increasingly important impact on research and clinical practice. The overall goal of this domain is to provide neuroscientists and clinicians with computational methods to automatically extract and analyze anatomical information from large cohorts of brain images and to assist neurosurgical procedures.
In the continuation of previous research and development efforts in Image Processing, the main project of the methodological teams of IFR 49 will consist in the creation of a new Centre for the analysis of large databases for neuroimaging studies. This Centre will be a two level structure.
- The acquisition of a high performance computing system will allow the methodologists of the IFR to push further the development of BrainVISA software platform. This software environment is distributed throughout the IFR to speed-up transfer of technology from the specialists of data analysis towards clinicians and neuroscientists. BrainVISA inner structure will be refined in order to give access transparently to the new computing system from any IFR team.
- In the context of the national Alzheimer initiative, The IFR will propose the creation of a large scale facility for analysis of large databases of neuroimaging data. This facility will be proposed as a service platform to any national team involved in the field, with a specific focus on clinicians. The idea is to provide clinicians with the possibility to trigger large scale neuroimaging studies. The center will provide support for the design and for the interpretation of the results, and take in charge quality control, organization of the data in a centralized database and processing using various pipelines including manual quality control. The center will integrate most of the standard tools available worldwide for image analysis and research tools developed in France. Hence, the centre will allow to integrate mature tools developed in national methodology laboratories, helping them to reach potential users. Finally, the center will push for the creation of a supervising council in charge of harmonizing the collaborations between teams in order to allow meta-analysis of the different databases.
The BrainVISA platform will expand further, integrating the new processing methods performed by the methodological teams of IFR (team NEMESIS, CRICM, Inserm U678, ENST, LNAO, MIRCEN). For instance, a project driven by MIRCEN will provide IFR teams with a brainVISA plug-in dedicated to rodent and monkey imaging. This module integrates the 3D reconstruction of histological data and their registration with MR images.
Another line of research will capitalize on a segmentation method of the hippocampus and the amygdala developed at GHPS . We propose to adapt this method to a longitudinal study of the deformation of these structures during the evolution of the pathology, and also develop quantification of shape features of the hippocampus for the classification of various pathologies (E. Gerardin, O. Colliot). These analyses will be applied to images acquired on 1.5 T or 3 T scanners. A joint project between Neurospin and GHPS will concern the acquisition and analysis of ultra-high resolution images of the hippocampus at 7 Teslas by designing new protocols for manual segmentation of the internal structures of the hippocampus and then deriving processing for segmenting these features automatically from the 7T images. A new project gathering teams from Neurospin and La Salpêtrière will address the potential of 7T magnet for the study of deep nuclei in Parkinson disease. The main topics are diffusion imaging and the potential of new contrasts for delineation of small nuclei.
Similarly, GHPS and Neurospin will look at multidimensional classification. Our purpose is to design methods to automatically classify subjects into different groups based on a high number of features. These features can originate from different brain areas, different imaging modalities, etc. Specifically, we propose to develop classification approaches based on kernel methods. This new methodology will be used to design automated methods to assist the diagnosis of Alzheimer’s disease at the prodromal stage, which is an important clinical challenge. In particular, we plan to design classification methods which integrate whole-brain anatomical information obtained from structural MRI, metabolic data obtained from PET as well as clinical and biological markers.
We will pursue the collaboration between UNICOG and NEMESIS for development of software for MEG and EEG processing (Brainstorm toolbox) and develop coordinate research on fusion methods between EEG and fMRI for estimation of the cerebral activation and its dynamics using new non linear regularisation schemes based on bayesian theory (LNAO, NEMESIS, INSERM U678). An extension of this research will concern the estimation of functional connectivity at different time scales based on such recordings.