erc chess
logo CHESS
Challenges in the Extraction and Separation of Sources

Workshop on challenges in multimodality.

 

Registration now open.


Registration to the workshop is free but mandatory.



Program of the workshop



 9h00 —  9h30Welcome and Introduction
Christian Jutten
 9h30 — 10h15Physiologically informed Bayesian analysis of ASL functional MRI data
Philippe Ciuciu
10h15 — 10h45Coffee Break
10h45 — 11h30Combining MEG and EEG for functional brain imaging under sparsity constraints
Alexandre Gramfort
11h30 — 12h15Linear and Non-linear subspace methods for multimodal neuroimaging
Felix Bießmann
12h15 — 13h30Lunch Break
13h30 — 14h15Source Separation for Multivariate Data Fusion: Applications in Medical Imaging
Tülay Adalι
14h15 — 15h00EEG-fMRI: To what extent can temporal sources be linked to spatial activations?
Maarten De Vos
15h00 — 15h30Coffee Break
15h30 — 16h15From imaging biomarkers to theranostic tools for brain pathologies: limits and challenges
Christian Barillot
16h15 — 17h15Round table debate





How to get there



By tram (from the train station or Grenoble centre)

Take tramway B toward Gières — Plaine des sports, and get off at stop Gabriel Fauré. Looking in the travel direction, GIPSA is in the building on your left-hand side (ENSE3, Grenoble INP). Detailed instructions can be found here.

The workshop will take place on the first floor of building D, room Mont Blanc. Indications will guide you.


For all other transport

Check How to get to GIPSA lab for detailed instructions.




Abstracts (in alphabetical order)


Source Separation for Multivariate Data Fusion: Applications in Medical Imaging
Tülay AdalιUniversity of Maryland, Baltimore County, United States of America

Joint blind source separation allows one to take full advantage of the statistical dependence across multiple datasets, and thus, compared to separate analysis of each dataset, it not only leads to better performance but also allows the identification of a wider class of signals. This talk reviews the fundamental results on independent component and vector analyses, and presents two models for feature-based fusion of multi-modal data. Advantages and disadvantages of each model are discussed along with the selection of a model for a given task. Finally, a number of key challenges for data fusion are probed to identify important directions for future research.




From imaging biomarkers to theranostic tools for brain pathologies: limits and challenges
Christian BarillotIRISA, Rennes, France

One of the major challenges in clinical neuroimaging is to detect quantitative signs of pathological evolution as early as possible in order to prevent disease progression, evaluate therapeutic protocols or even better understand and model the normal history of a given neurological pathology.
A particular challenge is to find correlations between brain structures at the morphometric, structural, metabolic or functional level through a large set of multimodal images. MRI is the premier means to study the human brain through various acquisition protocols. This presentation will illustrate this challenge through the use of novel metabolic or structural MRI sequences able to provide relevant information at the micro-structural level. Long-term perspective will be discussed on how these new imaging biomarkers can translate to theranostic tools.




Linear and Non-linear subspace methods for multimodal neuroimaging
Felix BießmannTechnische Universität Berlin, Germany

Subspace learning approaches to multimodal neuroimaging have become increasingly popular. These methods have in common that they are simple to implement and they can be solved efficiently (and often exactly) using basic linear algebra. We have found these methods to be useful in all stages of multimodal neuroimaging analyses, starting from basic preprocessing and artifact removal to integration of multiple modalities with complex spatiotemporal coupling dynamics. I will present examples for each of these stages with data obtained with invasive and non-invasive recordings and analyses that use linear and non-linear models as well as parametric and non-parametric approaches to model the coupling.




Physiologically informed Bayesian analysis of ASL functional MRI data
Philippe CiuciuCEA/Neurospin and Inria Saclay, France

ASL fMRI data provides a quantitative measurement of blood perfusion. In contrast to Blood Oxygenation Level Dependent signal, the ASL signal is a direct and closer to neuronal activity measurement. However, ASL data has a lower signal to noise ratio (SNR) and poorer resolution, both in time and space. In this work, we thus aim at taking advantage of the physiological link between the hemodynamic (venous) and perfusion (arterial) components in the ASL signal to improve the estimation of the impulse responses of the neurovascular system. In a Bayesian framework, a linearization of this link is injected as prior information to temporally regularize the regionwise estimation of the perfusion response function while enabling the joint detection of brain activity elicited by stimuli delivered along a fast event-related paradigm. All the parameters of interest in space and time as well as hyper-parameters are computed in the posterior mean sense after convergence of a hybrid Metropolis-Gibbs sampler. In this way, we aim at providing clinically relevant perfusion characteristics for the analysis of ASL data in low SNR conditions.

This work has been done by Aina Frau (PhD student) and Thomas Vincent (postdoc fellow) under the joint supervision of Florence Forbes (INRIA Grenoble) and Philippe Ciuciu (CEA/NeuroSpin &INRIA Saclay).




EEG-fMRI: To what extent can temporal sources be linked to spatial activations?
Maarten De VosUniversity of Oxford, United Kingdom

Two non-invasive techniques for the study of brain functioning, electroencephalography (EEG) and functional magnetic resonance imaging (fMRI), have complementary advantages in terms of spatial and temporal resolution. Recording EEG and fMRI data simultaneously became standard practice in many labs and literature on new ways to integrate the data from both modalities is continuously growing. However, a still unanswered question is to what extent features present in one modality are reflected in the other modality. Many methods rely on some form of unraveling data sources of one or both modalities. We will discuss recent work on a powerful extension of the jointICA algorithm to simultaneously extract sources from multi-subject EEG-fMRI data. Additionally, we will also present data from an EEG-based neurofeedback study with continuous fMRI scanning where there is a partial but clear dissociation between the findings in both modalities. Taken together, the results demonstrate clear advantages of multimodal data integration, while pointing out the limitations of the often-praised complementarity.




Combining MEG and EEG for functional brain imaging under sparsity constraints
Alexandre GramfortTélécom ParisTech, France

Electroencephalography (EEG) and Magnetoencephalography (MEG) are noninvasive techniques that allow to image the active brain at a millisecond time scale. EEG and MEG sample the electric potential and the magnetic field induced by neural activation. They therefore provide complementary views on the same electrical phenomena. In this talk I will first briefly review the physics behind MEG/EEG measurements before diving into the statistical and computational challenges arising when combining these modalities for source localization. I will cover the problem of spatial whitening that involves covariance estimation from a limited number of samples [0]. Models considered are shrinkage estimates as well as generative latent factor models (Bayesian PCA, Factor Analysis). Then I will detail some recent contributions on the inverse problem using structured sparsity promoting regularizations, time-frequency representations and the optimization solvers based on first order proximal methods as well as block-coordinate descent [1,2,3].

[0] D. A. Engemann, A. Gramfort, Automated model selection in covariance estimation and spatial whitening of MEG and EEG signals. To appear [1] A Gramfort, M Kowalski, M Hämäläinen , Mixed-norm estimates for the M/EEG inverse problem using accelerated gradient methods , Physics in medicine and biology, 2012
[2] A Gramfort, D Strohmeier, J Haueisen, M Hämäläinen, M Kowalski, Time-Frequency Mixed-Norm Estimates: Sparse M/EEG imaging with non-stationary source activations , NeuroImage 2013
[3] D Strohmeier, J Haueisen, A Gramfort , Improved MEG/EEG source localization with reweighted mixed-norms . Pattern Recognition in Neuroimaging (PRNI), 2014 http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=6858545&tag=1 http://hal.archives-ouvertes.fr/hal-01044748