Context and general aim
The extraordinary progress of neuroimaging, genomic and biomarker technologies has allowed to capture various characteristics of the brain disease development in living patients. Collection of multimodal data in large patient databases provide a comprehensive view of brain alterations, biological processes, genetic risk factors and symptoms. The team aims to build numerical models of brain diseases from multimodal patient based on appropriate data-driven approaches. To this end, we develop new data representations and statistical learning approaches that can integrate multiple types of data : neuroimaging, peripheral biomarkers, clinical and omics data (genetics, transcriptomics).
In particular, we develop methods to highlight networks of interactions among multiple sources of data, to track data changes during disease progression, and to automatically predict current or future clinical outcomes from these data. We apply these models to neurodegenerative disease (Alzheimer’s disease and other dementia, Multiple Sclerosis, Parkinson’s disease…). They shall allow to deepen our understanding of neurological diseases and to develop new decision support systems for diagnosis, prognosis and design of clinical trials.
New representations from multimodal medical images
Combining multiple neuroimaging modalities is necessary to obtain a comprehensive picture of alterations in brain diseases (atrophy, anatomical disconnections, functional connectivity alterations, metabolic alterations, abnormal protein deposits…). Such a combination is a non-trivial task because different types of information are conveyed by the different modalities, which in turns leads to different natural data representations (meshes and curves for geometrical information, signals, networks). We propose to build new integrated data representations from multiple modalities. Such representations will be subsequently entered into statistical models and decision support systems. Specifically, we will introduce representations that can integrate geometrical information (anatomical surfaces extracted from anatomical MRI, white matter tracts extracted from diffusion MRI) together with functional (PET, ASL, EEG/MEG) and microstructural information.
Network theoretic approaches to integrate heterogeneous brain networks
The complexity of biological systems often emerges from interactions between components at multiple spatial and temporal scales. Neglecting this information, and analyzing separately the levels of such scales, is an oversimplification of the real phenomenon. We propose a methodological framework that aims, on the one hand, to integrate information from networks describing different modes of connectivity (e.g. multi-modal, cross-frequency) and, on the other hand, to statistically model the organizational mechanisms of temporally dynamic networks (e.g. nonstationary, longitudinal). Target applications include: i) human learning in brain-computer interface, ii) prediction of neurodegenerative disease progression, and iii) identification of driving nodes in biological pathways (gene expression networks).
Spatio-temporal models to build trajectories of disease progression from longitudinal data
Longitudinal data sets are collected to capture variable temporal phenomena, which may be due to ageing or disease progression for instance. They consist in the observation of several individuals, each of them being observed at multiple points in time. The statistical exploitation of such data sets is notably difficult since data of each individual follow a different trajectory of changes and at its own pace. This difficulty is further increased if observations take the form of structured data like images or measurements distributed at the nodes of a mesh, and if the measurements themselves are normalized data or positive definite matrices for which usual linear operations are not defined. Our team has contributed to the definition of a generic theoretical and algorithmic framework for learning typical trajectories from longitudinal data sets. This framework is built on tools from the Riemannian geometry to describe trajectories of changes for any kind of data and their variability within a group both in terms of the direction of the trajectories and the pace at which trajectories are followed. The inference is based on a stochastic Expectation Maximization (EM) algorithm coupled with Markov Chain Monte Carlo methods.
Decision support systems for diagnosis, prognosis and design of clinical trials
Based on the new representations and statistical models, we design decision support systems for diagnosis, prognosis and design of clinical trials. These systems are based on: i) personalization of statistical models to predict evolution at the patient level; ii) new machine learning approches for classification and regression on high-dimensional and structured data; iii) content-based retrieval techniques.
University College London, UKUniversity of Utah, USAQueen Mary University of London, UKUniversity of MinnesotaINRIA Asclepios, FranceENS de Cachan, FranceUniversity Paris-Descartes, FranceUniversity Paul Sabatier, FranceInstitut Pasteur, FranceINSEAD, FranceCEA Neurospin, France
Sainte Anne hostpital, Paris, FranceBordeaux university hospitalCean University HospitalLille University Hospital
- IM2A / ICM Bruno Dubois’s team (Bruno Dubois, Harald Hampel, Marc Teichmann, Isabelle Le Ber)
- ICM Alexis Brice’s team (Alexis Brice, Isabelle Le Ber, Christel Depienne, Jean-Christophe Corvol)
- ICM Marie Vidailhet / Stéphane Lehéricy’s team (Marie Vidailhet, Stéphane Lehéricy, Andreas Hartmann, Yulia Worbe)
- ICM Catherine Lubetzki / Bruno Stankoff’s team (Bruno Stankoff, Benedetta Bodini)
- ICM Brahim Nait Oumesmar / Anne Baron Van Evercooren’s team (Violetta Zujovic)
- Department of Neuroradiology, Pitié-Salpêtrière Hospital