Magnetoencephalography Flashcards
What are we measuring with MEG?
- If enough cortex is activated this electrical potential difference can be measured with EEG
- Whenever there is current flowing through a medium it also produces a magnetic field (according to the righthand rule) through brain tissue without distorsion
Mag field is perpendicular to the current
Can measure whats going on in the head by measuring the mag flux, mag flux rapidly decays with distance but theoretically goes on forever, passes - The MEG signals derive from ionic currents flowing in the dendrites of neurons during neural activity
Mainly from pyramidal cells bc they’re well aligned with each other
If gr of neurons respond to the same kind of inputs, get summation of current and of mag field - Neural cells situated in the cortical surface of the brain have similar orientation and therefore, reinforce each other’s magnetic fields
To generate a signal that is detectable, approximately 50,000 active neurons are needed
Fundamental limit of spatial res
What do we measure MEG with?
To measure these minute signals, SQUID magnetometers are used
Superconducting Quantum Interference Device
Extremely sensitive magnetometer
Superconducting: reduce the resistance to as close to 0 as possible, sensors are bathed in liquid He
These SQUIDS become superconductors by using liquid helium to keep them close to absolute zero (minus 273.15 degrees Celsius)
Cooling allows for the special wires to change from normal conducting to super conducting (very little resistance) so the very small mag fields from the brain can create currents in them (magnetic induction)
Brain produces a current which has an associated mag field which hits conductors and creates a current that we can then measure
* As the signal and conductors are so sensitive, they must be hosted in a MSR (Magnetically Shielded Room)
What does an MEG machine look like?
- Magnetic field strength rapidly decays with distance, so the sensors must be close to the head
Have insulation layer between human and the liquid He (gap between head and sensors but gap has to be as small as possible so that signal doesnt decay too much between head and sensors)
What are the steps in a MEG study?
- Digitisation
- Set up participants
- Coregistration and head position measurement
Unlike with EEG, in MEG the sensors arent attached to the head so mvmt is a lot more of a prob, the sensors are in diff spots in relation to everyones brain, put coils on an inside cap and allow to localise where the person’s head in relation to the sensors and if they move, can track themvmt in relation to the sensors
What do raw MEG data look like?
- Time, Space, Frequency, Power and Phase
MEG data are multidimensional and thus provide many possibilities for hypotheses testing in neurophysiology/psychology .
Raw data looks a lot like EEG, have ltos of time series, each channel is relative to each sensor location Positive = source (mag field coming out, red), negative = mag field going back in head, blue Can also separate the signal in diff frequency bands Have many sensor locations in space, time, can have lots of frequency domains = 4 possible dimensions of data - The magnetic fields produced by the brain are very small compared to other sources (100 million times less than earth magnetic field)
Anytg in the env causing mag field causes artifacts
Biological signals: ex muscles in the face create mag fields, have to keep subject as relaxed as possible - The signal that you actually want in buried under a lot of noise
- We need several different ways to extract the signal from the noise
How do we pre-process MEG data?
- Down sampling (to reduce data size and subsequent data processing time)
- Artefact rejection (to remove noises that may contaminate brain signals)
- noises recorded from reference sensors * visual inspection
- signal processing techniques, e.g., linear transformations (PCA & ICA)
Have sensors that are mostly affected by outside mag fields, can subtract it from the data, same thing for mvmt and bio artifacts - Filtering (to remove low frequency fluctuations, high-frequency and power line noises)
Line frequency from electrical outlets (50 Hz alternating current) so can put a filter to take out 50 Hz
There shouldn,t be anytg of interest in the brain below 1 Hz so filter it out, same thing for above 100 Hz - Epoching (or segmenting) the long signal into smaller chunks based on events (e.g., when a stimulus was presented, when a movement was made etc)
If do smtg event-related, segment the data around the time stamps of the events
What should we bear in mind when reading an MEG paper?
- Preparation of subjects (use of non-magnetic clothes, questionaries to exclude participants)
- Instructions and debriefing of subjects
- MEG system: model, version, the acquisition and analysis software
- MRI: model, sequence, voxel size
Collect the shape of the head so can match the shape of the head to a structural scan, then project the fctional data (MEG) onto the structural MRI scan - Head-shape: How many digitized time-points?
- Coregistrtaion: How the individual head position recognized in the MEG system?
- Head movements: Were the head movements recorded? What was the max.?
Difference can take recording of head at begining and end and see how much it moved, if moved too much, exclude subject bc if there was activation, cant be sure where it was, with kids, take continuous recording of MEG and also of position of head so track mvmt between head positions before and after the run should be reported. - Position of participants: Was the participant in seated or supine position? How was the head of the participant positioned in the MEG Dewar?
- External recordings: State delays with respect to the MEG triggers. For example, a photodetector can be used to determine visual stimulus onset. Were the data corrected for the delay?
How do we analyse MEG data?
Sensor Level
- Event-related Fields (ERFs)
Change in timing of components, change in amplitude of components, or presence/absence of particular components
Take timing of events, chop up data and see if there was a response
- Spectral Analysis
Increase/decrease in power/phase in a particular frequency band locked to a stimulus (evoked oscillatory activity)
Increase/decrease in the power/phase in a particular frequency band not-locked to a stimulus (induced oscillatory activity)
Can decompose signal in frequency domain
Source Level
- Source Analysis– ‘the Inverse problem’ Reconstruction the neural sources underlying the signals measured at the sensor level- often combined with one of the above methodologies
Infer where signal came from in brain, have to know how mag field permeated in brain, mag field normally isnt affected by tissue
What is event-related fields (ERFs) ?
- Waveform deflections that have a significant amplitude with respect to the baseline of the recordings
- Assumption is that recorded brain signals are time locked and “phase-locked” to a specific event
We assume that they occur at the same time and same phase, if phase is diff, it’ll get averaged out, we assume that things at diff phases are noise - The rest of the waveform is considered “noise”
- Averaging multiple trials together keeps the evoked response while the noise is cancelled out
- Works well for early components, but later components tend to smear
What research questions you can ask with ERFs?
- Is a component present in one condition and absent in another?
- Is a component delayed in one condition relative to another condition?
- Does the size of a component vary with some feature of the stimulus? ex does amplitude of the signal vary with brightness or loudness of a signal
- Is a component present in one population and absent/reduced/delayed in another?
Caveat: ERFs vs ERPs
Scalp distributions and component polarities are different between ERPs and ERFs
For EEG, the positive and negative patches on an EEG scalp distribution lie at the end and beginning of the current source
For MEG, the positive and negative patches lie on either side of the current source
* MEG primarily picks up magnetic fields produced by current flows in the cortical pyramidal cells
* MEG optimally picks up current that flows tangential to the skull, i.e., from radial dipoles
* EEG detects current flows in pyramidal cells and is more sensitive to current flows deeper in the brain
* EEG optimally picks up radial current sources (i.e., tangential dipoles) but can see both type of sources
What are the different types of MEG sensors?
Differences between different types of MEG sensors *
There are two main types of sensors used in MEG
* Axial gradiometers: If signal comes from brain, itll more strongly activate the bottom sensor
* Planar gradiometers: measures change in flux over space
* Depending on the system, the data can look very little different
Differences can be found in: Especially the scalp maps, But also the waveforms And the units of measurement: ft vs ft/cm
What is a gradiometer?
Gradiometer = record from 2 sensors at
the same time and take the diff between
them
If get equal signal from both of them, the
signal is not coming from the brain, its
noise and it’ll be cancelled out
How should we represent our waveforms depending on the MEG sensor used?
Because of the difference between waveforms from axial gradiometer systems and planar gradiometer systems it is a good idea to present RMS waveforms because there are fewer differences in the RMS waveforms between systems making them easier to compare
There are some mathematical tranformations that enable you to represent your data in one way or another
How is grand averaging a problem in MEG?
- MEG head-sensor position can be very different across subjects
Sensors arent all in the same place for every subject
Cant make inferences about whats going on across the population like you can in EEG - Grand averaging is not recommended
- There are some method that can realign individual head-sensor coregistration to a common space to enable MEG grand averaging
Present stats in terms of source space, time signal aligned across all subjects based on where their heads were in comparison to sensors
