Neurophysiological significance Neurophysiological significance of the inverse problem of the inverse problem its relation its relation to present “source estimate” to present “source estimate” methodologies methodologies and to future developments and to future developments E. Tognoli Discussion group about Source Estimation 5 November 2004
34
Embed
Neurophysiological significance of the inverse problem its relation to present “source estimate” methodologies and to future developments E. Tognoli Discussion.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Neurophysiological significance of the Neurophysiological significance of the inverse probleminverse problem
its relation its relation to present “source estimate” to present “source estimate”
methodologies methodologies and to future developmentsand to future developments
Neurophysiological significance of the Neurophysiological significance of the inverse probleminverse problem
its relation its relation to present “source estimate” to present “source estimate”
methodologies methodologies and to future developmentsand to future developments
E. Tognoli
Discussion group about Source Estimation
5 November 2004
I. A “poor spatial resolution”I. A “poor spatial resolution”I. A “poor spatial resolution”I. A “poor spatial resolution”
EEG : a poor spatial resolution EEG : a poor spatial resolution
• Priors :
– We are repeated that EEG has a “poor spatial resolution”, although good temporal one : thus, assumptions on structure-to-function are loose
Let’s localize the sourcesLet’s localize the sources
• Each of these 3 topographic maps comes from a single dipole activation of the cortex in a dipole simulator program.
Estimate the locations of these 3 single sources
Let’s localize the sourcesLet’s localize the sources
Why?Why?II. What distorts the signal?II. What distorts the signal?
Why?Why?II. What distorts the signal?II. What distorts the signal?
Volume conductionVolume conduction
• Principe of EEG recording: volume currents
• What you want to know: where (projected on scalp) something is happening
• What you get: a large ususally bipolar propagation of the “source”
• Effect : Blurs the signal
AnisotropyAnisotropy
• Anisotropy : density/impedance of tissues
• distort the topography of the signal recorded from the scalp, as compared to the signal recorded from the cortex
• Act as a spatial low-pass filter
• Effect : Blurs the signal
More of anisotropyMore of anisotropy
• The sinuses: partially filled with… nothing
• Effect : displaces the signal
Brain foldingBrain folding
• Brain folding : the EEG signal mainly originates from pyramidal layers III & V.
• The orientation of the active patch of neuron is of prime importance for the projection of the activity onto the electrodes
• Blurring of the source – Anisotropy– Volume conduction
• Displacement of the source– Cavities – Brain folding
• The signal is corrupted both in its extent and in its location
III. Some solutions ?III. Some solutions ?III. Some solutions ?III. Some solutions ?
Some solutions?Some solutions?
If pretending to do some topographical analysis of the EEG:
• Because of this corrupted correspondence between the sources of bioelectrical activity and their scalp topography, we are lead to work, not in the space of the electrodes (maps/splines of raw signals), but in the space of the currents (maps/splines of estimated sources of raw signal)
• We do not record directly the cortex,but do as if, with a mathematical reconstruction
Some solutions?Some solutions?
• Solutions has been proposed through either :– Mathematical transform (eg. Laplacian, CSD)– Estimation/modeling (minimization of Laplacian
in 3D (voxel-based) models : inverse problem)
Some solutions?Some solutions?
• Solutions has been proposed through either :– Mathematical transform (eg. Laplacian, CSD)– Estimation/modeling (minimization of Laplacian
in 3D (voxel-based) models : inverse problem)
Some solutions?Some solutions?
• The Laplacian is problematic for spatial analysis of EEG data (eg. coherence analysis), since it projects correlated activities in 2 (presumably silent) unrelated locations of a tangential foci : source and sink
• Although these data can be correctly interpreted (at least for a few sources), they lead to incorrect assumption of the structure-to-function relationship in the literature
Some solutions?Some solutions?
• Solutions has been proposed through either :– Mathematical transform (eg. Laplacian, CSD)– Estimation/modeling (minimization of Laplacian
in 3D (voxel-based) models : inverse problem)
Some solutions?Some solutions?
• Estimation of sources : 2 approaches– Dipole: one single (or a few) point-like sources,
center of mass of localized activity) : not useful for spectral/coherence analysis : information is excessively reduced
– Smooth current estimates (reconstruction of time series at many (N’>N) spatial locations by estimating solutions to the inverse problem)
Some solutions?Some solutions?
• No unique solution to the inverse problem– Under-determination (ill-posed problem)
• Two crucial points for the accuracy of the estimation– Performance/assumptions of the algorithm (given an
undetermined, noisy signal)
– Accuracy of the head model
IV. AlgorithmsIV. Algorithmspresent methodspresent methods
IV. AlgorithmsIV. Algorithmspresent methodspresent methods
AlgorithmsAlgorithms
• Inverse problem with smooth solution– MN (or MNE - minimum norm estimate, or LE, linear estimation)
(Hamalainen & Ilmoniemi, 1984) – WMN (weight the contribution of sources regarding depth, to rule
out the bias toward high contribution from sources close to the surface)
– LORETA (low resolution electromagnetic tomography)- generalized weighted minimum norm/laplacian : extend the properties of MN by projecting solutions in true 3D (VOXEL BASED)(Pascual-Marqui)