Please enable JavaScript.
Coggle requires JavaScript to display documents.
Task 4: MEG, EEG, oscillations and sources (CM (A (right hand rule thumb…
Task 4: MEG, EEG, oscillations and sources
(1) Gazzaniga (3rd) - CN - Chapter 3 - Methods of CN pp148-152
Functional Imaging
Electrical and Magnetic Signals
- ERP in visual area to identify multiple sclerosis
- AEP to identify tumors in auditory pathways
- early AEP localized with the help of other methods, late components cannot be localized
Localization
- forward solution: demonstration how a specified dipole would cause a certain EEG measurement pattern, based on simplyfied assumptions of the sphericity and homogeneity of the brain
- inverse problem: based on measured EEG activity we cannot infer the location of the dipole/dipoles that caused it, because there's an infinite number of possible solutions
- one assumption in inverse modelling is that neural generators can be modeled as electrical dipoles, conductores with one positive end and one negative end
- we guess a location of a dipole, produce the forward solution and compare to recorded data
MEG
- undistorted signal, thus better fit for localization
- most of the MEG signals stems from intracellular current flowing within the apical dendrites of pyramidal neurons
- can only measure neurons that are oriented tangential
(2) Hari, R. & Lounasmaa, O.V. (2000). Neuromagnetism: tracking the dynamics of the brain
pp33-38 (full doc) this articles states that it is "weak electrical impulses transmitted between brain cells" (so if the current flows through the dendrite, is that a signal "between" neurons?) MEG: How it works
- superconducting quantum interference devices (SQUIDS)
- superconducting flux transformer
- sensor units
- gradients coils (deriv. of y,z of B) - magnetometer coil (z of B)
- insulation of layer of aluminium and mumetal
this portion of the text really doesn't explain these terms any further
Neural Currents underlying MEG signals
- simultaneous & equally oriented activity creates a current that is "dipolar" in character
- non-uniqueness problem
Calculating the sources of the magnetic fields
- assumptions (spherical, 1 or more dipoles
- the fields from all other current cancel each other out because the brain is approximately spherically symmetrical
- multidipole models ( multiple dipoles being active (nearly) simultaneously
- minimum current estimate???
Applications of MEG
- spatial accuracy of typically a few milimetres
Cortex-muscle coherence
Action viewing
- suppression of 20 Hz activity in motor cortex during own movement, imagined movement and observation of movement of a different person
Clinical applications
- identification of epilepsy origin sites and tumor locations
- combination with MRI data (surgeon can circumvent blood vessels)
MEG versus other techniques
- some current from deep within the brain are better picked up by EEG
The future of MEG
- if SQUIDS could be developed that work under higher temperatures, the insulation could be thinner, and placement more close to the scalp, creating better SNR would be possible
-
(3)Luck, S. J. (2005). In: An introduction to the event-related potential technique Chapter 7. ERP Localization. pp. 267-301
- explain ERP source localization and provide a discussion and critique of the most common source localization techniques.
The Big Picture
- once several dipoles are present, no unique solution anymore
- One approach is to use a small number of (A) equivalent current dipoles, each of which represents the summed activity over a small cortical region (perhaps 1–2 cm3), and assume that these dipoles vary only in strength over time; this is the equivalent current dipole category of source localization methods
- The second category divides the brain’s volume (or the cortical surface) into a fairly large number of voxels (perhaps a few thousand), and computes the set of strengths for these voxels that can both explain the observed distribution of voltage over the scalp and satisfy additional mathematical constraints; this is the (B) distributed source category.
The Forward Solution
spheres to finite element models
- determine resistance of individual cubicles via MRI (tissue types)
boundary element model: focuses on boundaries of tissue and assumes constant resistance throughout one tissue type
(A) Equivalent Current Dipoles and the BESA Approach
- <10 dipoles assumed, with fixed location and orientation, and magnitude changing over time
- once this is set, magnitude for best fit is computed
- algorithm moves adjustment (loc,orie,mag) towards better fit
The Essence of BESA
The Starting Point
- How many dipoles? What starting position?
- principal component analysis (gives #)
- starting with low number of dipoles that explain first part of signal, adding dipoles from there on to explain later parts
- multi-start approach: which locations occur independent of starting conditions?
Shortcomings of the BESA Approach
- high operator dependance
- no way of quantifying error
A Simulation Study
- simulation study by Miltner et al. (1994)
- Despite the fact that the simulation perfectly matched the assumptions of the BESA technique and was highly simplified, none of the participants reached a solution that included all ten dipoles in approximately correct positions
- under the simplified conditions, dipoles were often localized with a reasonable degree of accuracy, however many were missed, merged, and some spurious dipoles were added
(B) Distributed Source Approaches
General Approach
- parcellation into voxels, each with individual dipole, from here same approach as above: give values, compare forward solution to observed data
Cortically Constrained Models
- reduces number of free parameters
- is this a subtype of the general approach or sth same?
The Minimum Norm Solution
- Ha¨ma¨la¨inen and Ilmoniemi (1984) proposed selecting the one solution that both produces the observed scalp distribution and has the minimum overall source magnitudes.
- depth-weighted
- LORETA technique uses smoothness constraint (now can only find center of an area of activation)
- even adding constraint based on other imaging data may not improve solution: fMRI data may show things that EEG does not and vice versa, but using fMRI based constraints will bias localization towards fMRI based activity
The Added Value of Magnetic Recordings
- combining ERP and ERMF data provides a new set of constraints that can aid the localization process
- drawbacks: (a) a more complex head model is needed for the electrical data, and (b) some effort is required to ensure that the electrical and magnetic data are in exactly the same spatial reference frame.
Can we really localize ERPs?
Source Localization as Model Fitting
- need margin of error
- However, the margin of error that could be specified in this manner would be meaningful only if the constraints of the model were fully adequate and the only sources of error arose from noise in the ERP data (and perhaps errors in specifying the head model).
Probabilistic Approaches
- report not one but many solutions that fit criteria
- if one area is includes in various solutions, employing different criteria certainty that this area is involved would increase
- a distributed source localization technique based on Bayesian inference that provides a more sophisticated means of assessing probabilities.
Recommendations
- only experts, only with biological constraints, see only as piece of converging evidence
Source Localization and Scientific Inference
In this context, a given source localization model will have value to the extent that it not only supports a specific hypothesis but is also unlikely to have been obtained if the hypothesis is false.
- until margin of error, localization will only be converging evidence
Specific Recommendations
-
(4) Pizzagalli, D. A. (2007). Electroencephalography and high-density electrophysiological source localization (3 Ed., pp. 56-67)
1. INTRODUCTION
- limit to number of electrodes that can be used
2. Physiological basis of the EEG
2.1 EEG generation: 1. The role of the post-synaptic potentials in cortical pyramidal neurons
- far-field potentials
2.2. EEG generation: II. The role of thalamo-cortical networks
- cortical alpha power was found to be inversely correlated with glucose metabolism in the thalamus
- In sum, EEG oscillations appear to be dependent on interactions between the cortex and thte thalamus, which both produce intrinsically rhythmical activities.
2.3. EEG generation: III. The role of local-scale and large-scale synchronization
- emerging evidence indicates that oscillatory processes might be implicated in many cognitive processes
- higher frequency oscillations appear to originate from smaller neuronal assemblies, whereas low frequency oscialltions span larger neuronal populations
- EEG coherence analysis
3. NORMATIVE EEG ACTIVITY
- allows to differentiate between functional inhibitory and excitatory activities
3.1. Delta band (1–4 Hz)
- Delta current density??
- sleep, pathology (lesions)
- predominant activity in infants
- sum: suggests inhibitory function
3.2. Theta band (4–8 Hz)
- sleep, drowsiness, but also focused attention (in frontal area) + another form related to mental effort
3.3. Alpha band (8–13 Hz)
- relaxed wakefulness
- alpha as a sign of cognitive idling
- sub-bands?
3.4. Beta band (13–30 Hz)
- increases during attention
3.5. Gamma band (36–44 Hz)
- associated with large-scale integration
4. DATA ACQUISITION AND SIGNAL ANALYSIS
4.1 Electrodes
4.1.1. Electrode locations and high-density recordings
- Studies using simulated as well as real EEG data have suggested that an electrode distance of 2.3cm is required to prevent distortions of the scalp potential distribution, and thus allow resolution of spatially focal EEG patterns.
- both simulation and experimental studies suggest that at least 60 (preferably more) equally distributed electrodes are required for accurate spatial sampling of scalp activities.
4.1.2. Electrode interpolation
- how to deal with bad/ missing data from one electrode
- not possible to just omit data
- weighted average of neighboring electrodes
- non-linear spline interpolation
4.1.3. Recording reference choice
- close reference will flatten signal since close areas frequently show synced activity
- in particular, the use of an average reference (Lehmann, 1987), radial current flow (Hjorth, 1975), and current source density (Perrin, Bertand, & Pernier, 1987) have attracted substantial interest.
4.2. Recording: Filters and sampling rate
- Nyquist Theorem & spurious low frequency components
4.3 Artifacts
- regression approaches to deal with artifacts?
- independent component analysis
5. QUANTITATIVE SCALP ANALYSES
- coherence?
- assumption of EEG signals as stationary processes ?
5.1. Spectral analyses
- length of EEG samples determines lowest possible frequency to be analysed (a 2sec sample cannot be used to analyse frequencies below 0.5Hz)
5.1.1. Asymmetry metrics
5.2. Time-frequency analyses(PART OVER) -
(3) What does it tell us?
- about the neural basis
- about functional aspects of psychology
- working mechanism of MEG
- what are event-related fields?
- different types of oscillations?
- anything else than state of excitation?
- Fourier transformation
CM
A
right hand rule
- thumb point towards negative pole, thus opposite to the direction of the electron flow
- the magnetic field is circular around the wire, thus cylindrical in total
B
oscillation frequencies
- What can we say about the development of amplitude and frequency from coma to excitation?
- if the amplitude goes down, that means that there's less synchronized activity, right?
- so, the "default" pattern of cortex activation (no! of pyramidal cells, perhaps also of the rest of the cortex) is synchronized activity??
Spectral Analysis / Fourier Transformation
- all frequencies need to be present across the whole sample
- absolute power
- relative power: (look that up!)
C
localization
- How complex is this task?
- How big of a problem is it?
MEG in localization
- How does the fact that the MEG signal is not distorted lead to MEG being helpful in localization?
- How does the fact that MEG can only measure current that are tangential to the scalp surface, affect localization?
-
-
-