Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Participating TeamsVisages U746
HybridPanama
EA 4712 (associated)Athena (associated)
HEMISFERHybrid Eeg-MrI and Simultaneous neuro-FEedback
for brain Rehabilitation
Start date: 01-2014
General objectives of HEMISFER
• Make full use of neurofeedback (NF) paradigm forbrain self-regulation/stimulation in:• Rehabilitation (ADHD, Strokes, …)• Psychiatric disorders (anxiety, schizophrenia,
resistant mood disorders, ….)• Main Challenges:
• Learn a coupling model associating functionaland metabolic information from simultaneousMagnetic Resonance Imaging (fMRI) andElectro-encephalography (EEG)
• Enhance the NF paradigm from the couplingmodel
Context
From Imaging Biomarkers to Therapeutic Concept
EEG
Neurovascular
coupling
ASL
BOLD
fMRI
Neuronal activity
Cerebral blood flow and volume (CBF, CBV)
BOLD Signal
Energy needs of brain tissues
Consumption of glucose (CMR_Glc)
& O2 (CMR_O2)
Blood concentration in dHb
Local magnetic perturbation due to dHb
Example :Perfusion abnormalities in chronic resistant depression[Batail JM et al. 2014]
Group Analysis
Right Hippocampus + Amygdala
(p<0,02 FWE)
Resistant Group > Control Group (n=7)
R+L Hippocampus, R median Cingulum
(p<0,05 cluster FWE)
Resistive Patient 16 > control group
(n=30)
Resistive Patient 01 > control group
(n=30)
Hippocampus + Amygdala(p<0,05 cluster FWE)
Individual Analysis
Data Processing
• Pretreatments (artifacts, …)
• Real time data analysis
• Extraction of patterns of interest
Production of
« feedback »
HEMISFER
Project
Current Status:• fMRI with Siemens MR 3T scanner• EEG with Brain Products 64-channel system• Real-time EEG and fMRI data acquisition,
processing, NF calculation and visualization infull synchronization.
Future Challenges:• Online NF with joint modeling approaches• Improve performance with GPU based real-time
parallel processing
Experimental Platform
Major Challenges
• Develop new NF paradigms able to profit from simultaneous EEG/fMRI/fASL
recordings
• We expect these novel paradigms to be able to concentrate the brain
metabolism on specific regions of the brain
• Learn models at the signal level able to explain the coupling of EEG and fMRI
signals under simple and more advanced brain stimuli (e.g. BOLD fMRI, fASL,
basal ASL)
• Learn both the domain in which brain activity is sparse (e.g., dictionary
learning), and adjust parametric models of the acquisition processes
• Achieve super-resolution in the spatial and frequency domain by
expressing the problem as a linear inverse problems regularized with the
learned coupled model
• Use brain connectivity models as prior information (later stage)
• Use the learned coupling models in order to “enhance” the EEG signal while
performing the same stimuli and NF tasks outside MRI
Coupling Model
Introduction of a new linear EEG/BOLD coupling modelusing sparsity-based regularization term:
Participating People : Elise Bannier, Christian Barillot, Jean-Marie Batail, Isabelle Bonan, Maureen Clerc, Isabelle Corouge, Dominique Drapier, Remi Gribonval, Anatole Lecuyer, Jussi Lindgren, Marsel Mano, Saman Noorzadeh, Pierre Maurel, Thomas Oberlin, Lorraine Perronnet
Simultaneous EEG-fMRI data acquisition Off-MRI EEG data acquisition
Learned EEG-MRI coupling model
BCI program
Visual Stimuli
Data Processing
Patient
BCI program
Visual Stimuli
Patient
Data Processing
Model Update
Coupling Model
Paradigm
EEG signal leadfield mixing matrix fMRI signal Hemodynamic model
Results
In the first study:• We compared EEG-NF, fMRI-NF and EEG-fMRI-NF
on a motor imagery task with 10 subjects• We propose an integrated feedback metaphor for
EEG-fMRI-NFSparse regularization
fMRI Subsystem
NFB Unit
EEG object
fMRI object
RecorderRecView
Buffer
RDA
RDA
Visualize
NFB Control Joint NFB
RDARDA
USB
Parallel
HDMI
USB
EEG Subsystem64 Ch. EEG Cap
2 x 32 Ch. Amplifiers
USB adapter
Brain Activity
Brain Activity
Trigger Unit
MR Head Coil
MR Console
HDD
Buffer
Cues/NFB
Display
SyncBox
*Recorder and RecView are commercial software, the rest of the NFB UNIT modules are build in-house (Matlab/C++)
sync &control
Patient
Acquisition of brain functions (MRI/EEG)
• Motor imagery-related fMRI activations were stronger during EEG-fMRI-NF than EEG-NF
• During EEG-fMRI-NF participants regulated more the modality that was harder to control
In the second study, we compared 1D EEG-fMRI-NF and 2D EEG-fMRI-NF
localization based on EEG localization based on fMRI
localization based on EEG-fMRI three-layer head model