M/EEG Analysis

This section will provide you with information and resources to perform M/EEG analysis. We will mostly highlight the different analysis commonly performed by lab members and links towards relevant resources for them.

Preprocessing

Given that most members use MEG, we will focus on the analysis of this type of data. However, keep in mind that it will be fairly similar to EEG. If you want more basic information about M/EEG analysis in general, software to use or other resources check the MEG Resources page.

Fieldtrip is the default software being used for most M/EEG given that it is developed at the Donders and they provide support for it. Hence, most of the analysis described below will have links directing you to tutorials using this software. If you want alternative to this, please refer to the MEG Resources page.

For the preprocessing of MEG data, we follow several general steps:

  1. Load the raw participant data
  2. Define the trials (usually based on the triggers of your experiment code) – use ft_definetrial
  3. (Optional) Remove any trials that should be excluded from your main analysis (e.g., practice trials, incorrect trials, or trials from a specific type)
  4. (Optional) 3rd order gradient correction (use ft_denoise_synthetic): this helps to remove noise from a source outside of the brain/head and possibly from the environment
  5. Demean data (use ft_preprocessing with cfg.demean = ‘yes’)
  6. Reject trials with high MEG variance (use ft_rejectvisual and select trials manually) – main purpose is to remove SQUID jump artifacts
  7. (Optional) Reject trials with high muscle activity
  8. (Optional) Resample data at a lower sampling frequency to reduce data size (use ft_resampledata) – Note, resampling can change the time axis in undesired ways, in many cases it can be useful to manually define a time axis to which to resample your data so that the data are centered at zero
  9. Run ICA analysis to identify and reject components related to eye movements and heart beat (use ft_componentanalysis)
  10. Save preprocessed and cleaned data to disk

In the following link you can find a series of Fieldtrip based tutorials to perform all of the preprocessing steps mentioned.

Overview of Analysis Type

Analyses of MEG data can broadly be divided into sensor-level and source-level analyses. The former are done on the level of the MEG sensors, while the latter use some sort of source reconstruction technique to estimate activity and specific source locations.

Sensor Level Analysis

A major goal when analyzing M/EEG data is to examine the modulation of brain activity linked to a specific event. However, due to the large amount of intrinsic and extrinsic noise in the signals, it is nearly impossible to get any data from single trials. A very common approach is to repeat events in your experiment and average the corresponding neural signals over several trials, thus increasing your singal-to-noise ratio (SNR). The main assumption is that the noise is independent from the events and thus reduced when averaging, while the effect of interest stays in place.

The following links will direct you towards lectures and tutorials on how to perform these analysis for M/EEG data

Time Frequency Analysis

Oscillatory components contained in the ongoing EEG or MEG signal often show power changes relative to experimental events. These signals are not necessarily phase-locked to the event and will not be represented in event-related fields and potentials. The goal of this section is to compute and visualize event-related changes by calculating time-frequency representations (TFRs) of power.

The following links will direct you towards lectures and tutorials on how to perfrom these analysis for M/EEG data

Source-Level Analysis

The anatomical MRI is preferred if you want to do source modelling of the MEG signal. Source modelling (also known as “inverse modelling”, “source reconstruction”, “beamforming” or “dipolefitting”) requires that you have a volume conduction model. This volume conduction model (also known as “head model”) describes the relation between sources in the head and the measured activity on the outside of the head (either EEG or MEG), and it models the geometry and conductive properties of the head.

Ask Paul Gaalman for the standard MRI procedure and sequence.

The following links will direct you towards lectures and tutorials on how to perform these analysis for M/EEG data

Additional Analysis

MVPA