MIND
Graphs are used in modern neuroscience to represent structural and functional connectivity of the brain acquired using magnetic resonance imaging. A number of pathologies affect these networks: their study holds vast promise for understanding and diagnosing brain diseases. But brain connectivity data is not trivial to process, and the predominant graph-theoretical analysis approach is used mostly at the group level. This project aims at developing imaging markers of neurodegenerative diseases based on recognition of functional and structural connectivity patterns of the brain at rest, for use predictively at the single patient level. The methods will be complementary to existing imaging modalities and clinical practice. The project will address two important tied questions: 1) How can we extract connectivity graphs from neuroimaging data, and perform inference on such graphs? 2) What are the confounds in the subjects or data processing that may impede clinical application of the methods? These questions will be addressed by new theoretical work and experiments with data from early multiple sclerosis and Alzheimer's disease patients. Data preprocessing and estimation of functional connectivity will be evaluated and improved, new algorithms to compare functional and structural connectivity graphs will be developed, and novel multimodal machine learning techniques will be developed to leverage the structural/functional connectivity interaction. Differences in connectivity due to physiological noise, pathological changes, and scanning site will be modelled in depth, and methods for compensating for such confounding factors will be evaluated. The project will advance novel imaging markers with high clinical relevance for two very prevalent neurological diseases in the EU. The project's dissemination stance will foster EU excellence in the emerging science of connectomics, and the fellowship will equip the researcher with skills vastly improving his scientific standing.
Universite De Geneve